search

Oneflow-Inc / CoModels

3 stars 2 forks source link

模型适配进度（拓展） #163

Open kokuro-asahi opened 12 months ago

kokuro-asahi commented 12 months ago
总计：8+14+9+26+21+10+3+6+6+10+3+9+11+4+88=228 领域 功能 基础模型 支持方式 负责人 状态 展开数量 Onelab负责人 OneLab公开项目链接
cv classification EfficientNet_b0 flowvision ke 完成 8 li https://www.oneflow.cloud/drill/#/project/public/code?id=fa5438d8c14b8fa64429da52ea3aaa4b
cv classification EfficientNet_b1 flowvision cui 完成 8 li https://www.oneflow.cloud/drill/#/project/public/code?id=0919cef11021fa3729d9678b27bb5434
cv classification EfficientNet_b2 flowvision cui 完成 8 li https://www.oneflow.cloud/drill/#/project/public/code?id=fa8c4ace23affba27863ea9f56bbb662
cv classification EfficientNet_b3 flowvision cui 完成 8 li https://www.oneflow.cloud/drill/#/project/public/code?id=9486b2d7f8523973daf7b90d7fb9fc17
cv classification EfficientNet_b4 flowvision cui 完成 8 li https://www.oneflow.cloud/drill/#/project/public/code?id=d33198119b667ea422696576e5c67a4e
cv classification EfficientNet_b5 flowvision cui 完成 8 li https://www.oneflow.cloud/drill/#/project/public/code?id=d4c9a2d6db1a934b3896b56b92a03e74
cv classification EfficientNet_b6 flowvision cui 完成 8 li https://www.oneflow.cloud/drill/#/project/public/code?id=f9d51026c6d13eaca244a4d4c3eca1c3
cv classification EfficientNet_b7 flowvision cui 完成 8 li https://www.oneflow.cloud/drill/#/project/public/code?id=9b2d0d5ad34bf2eceff55f77007a260d
cv classification regnet_y_400mf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=74db0fcbf0b7b42d74a5c37d1dcd8c8d
cv classification regnet_y_800mf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=7af27966f711f8d0cf07b51d40a57992
cv classification regnet_y_1_6gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=579f7eb5c7d26076962e0721d1d709f9
cv classification regnet_y_3_2gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=5affa8b3893616466b0242397924a247
cv classification regnet_y_8gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=a38f54a87436578dafa7ac15927a9af9
cv classification regnet_y_16gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=35607cd00b6c22f5fd02254818b400ae
cv classification regnet_y_32gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=dfb022ad6b5b7e3b431c84e548dd53fa
cv classification regnet_x_400mf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=ed0d790aa2d1e911a72daa7e6605a97c
cv classification regnet_x_800mf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=d0097d4399eac0f6c31fe3dc29dacb17
cv classification regnet_x_1_6gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=9b2b5d00b3ff83198dfdc48a9085deb0
cv classification regnet_x_3_2gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=240b02c57ab1c27e98024acaf3aef70a
cv classification regnet_x_8gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=4e76ee1a1129ab9c2c5f903e6ffb7d39
cv classification regnet_x_16gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=36bca178cb147c64c3251ef38082c593
cv classification regnet_x_32gf flowvision cui 完成 14 li https://www.oneflow.cloud/drill/#/project/public/code?id=4b9bb2aa6ad9857fb31c96299fce9051
cv classification rexnetv1_1_0 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=f8e263b706369a6846c13985edad8433
cv classification rexnetv1_1_3 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=5acd5c7d3dcffd712779101d3af9e0ba
cv classification rexnetv1_1_5 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=edc766c2407bb08ca908516322282a19
cv classification rexnetv1_2_0 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=7e28ee4f9157562fa795899038da2a6e
cv classification rexnetv1_3_0 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=44f0cd0f722917ff90e921412d2fd775
cv classification rexnet_lite_1_0 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=8c0935ce218c7fe14721a54de380cf95
cv classification rexnet_lite_1_3 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=44a6598a4f7ce345a5d7c3ba458eb7e4
cv classification rexnet_lite_1_5 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=bc98489e6cc245e8348dd3289d237f97
cv classification rexnet_lite_2_0 flowvision cui 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=24f5e35e6056919730c4cabd354144c3
cv classification vit_tiny_patch16_224 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=25334d0e04ca945c25b331b67d2eb550
cv classification vit_tiny_patch16_384 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=e1302a0101857135ab62132d1e35aeae
cv classification vit_small_patch32_224 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=2648e2213ad95a43b35631ca58e0be84
cv classification vit_small_patch32_384 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=12768b29f94c5be820d11e9d5a3ae81f
cv classification vit_small_patch16_224 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=281132f74d7608078277ee4b46ef1701
cv classification vit_small_patch16_384 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=3f579f7f1350223ea18797a1a204c1e8
cv classification vit_base_patch32_224 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=18e3f3be24228e68ec8abaf3c51d1a08
cv classification vit_base_patch32_384 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=c76367e43afccb54509ef57a47641a2b
cv classification vit_base_patch16_384 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=5ef00487de1c66d4b59056aabe95e92a
cv classification vit_base_patch8_224 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=9b637c392fe656a10c5401cb205e5a1e
cv classification vit_base_patch16_224 flowvision 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=ae6f744c9a856d2f04784742f9161143
cv classification vit_large_patch32_384 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=96c37fd309acf0c9a007753e0782a181
cv classification vit_large_patch16_224 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=73350d451067d9a7164cf5f60f2915cd
cv classification vit_large_patch16_384 flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=20efdf3e5ce3526d180e9d68f03c533a
cv classification vit_base_patch16_224_sam flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=be4351753e1262778745eb1098c49c57
cv classification vit_base_patch32_224_sam flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=c73d578cc1f4059e05e7367b69c69bf7
cv classification vit_tiny_patch16_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_small_patch32_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_small_patch16_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_base_patch32_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_base_patch16_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_base_patch8_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_large_patch32_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_large_patch16_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_huge_patch14_224_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_base_patch16_224_miil_in21k flowvision zhang 没有ImageNet 21k数据集 26
cv classification vit_base_patch16_224_miil flowvision zhang 完成 26 li https://www.oneflow.cloud/drill/#/project/public/code?id=21423b30bd3ea7e717c486c5c96fabe8
cv classification deit_tiny_patch16_224 flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=13cfd9108b2f72cb92a99a846e98e6ff
cv classification deit_base_patch16_224 flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=af5947962e7720e97a570703a9e53694
cv classification deit_base_patch16_384 flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=606e242389cccb930249be457130dbf3
cv classification deit_tiny_distilled_patch16_224 flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=5891ddd5f03d6af0b2aeb690ced37b5d
cv classification deit_small_distilled_patch16_224 flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=849921cbc3a4c0b31f22203c0b60c690
cv classification deit_base_distilled_patch16_224 flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=9a58d5387296f5673265405938ec8d65
cv classification deit_base_distilled_patch16_384 flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=b1ebb442d627b9350681e2c8c97c74c2
cv classification deit_base_patch16_LS_224 flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=d17491afb893e3d0ab7915f53fcb674f
cv classification deit_base_patch16_LS_224_in21k flowvision zhang 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=425be3891ca118082c05635fd2305914
cv classification deit_base_patch16_LS_384 flowvision cui 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=ff8e43ef6752f38d46c6575296d244a6
cv classification deit_base_patch16_LS_384_in21k flowvision cui 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=6f1deac6410a3db19d7d22befba4fc0b
cv classification deit_huge_patch14_LS_224 flowvision cui 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=8a5517f5f84c5f6a2497c7307c1e6899
cv classification deit_huge_patch14_LS_224_in21k flowvision cui 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=8a5517f5f84c5f6a2497c7307c1e6899
cv classification deit_large_patch16_LS_224_in21k flowvision cui 进行中 22 li
cv classification deit_large_patch16_LS_384_in21k flowvision cui 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=274caa0440f9df12e989678ed205c0d4
cv classification deit_small_patch16_LS_224_in21k flowvision cui 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=30fbf529f1d61ac77b72514f8e913cc8
cv classification deit_small_patch16_224 flowvision li 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=ffabc32455e8966ad33bb07974f194bf
cv classification deit_small_patch16_LS_384 flowvision cui 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=e2e3159b7bcbd13a91e7736ee6ea3960
cv classification deit_small_patch16_LS_384_in21k flowvision cui 进行中 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=9be2c26a99c860a3f98c3f0a5b854e61
cv classification deit_large_patch16_LS_224 flowvision li 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=37ff4dabe9c2b9132e76badcabce7506
cv classification deit_small_patch16_LS_224 flowvision li 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=21f5789060c865a30c1eec1aade0c2c1
cv classification deit_large_patch16_LS_384 flowvision li 完成 22 li https://www.oneflow.cloud/drill/#/project/public/code?id=457dd8d362220fe32d340bfee7b7f306
cv classification mlp_mixer_s16_224 flowvision ke 无模型 10
cv classification mlp_mixer_s32_224 flowvision ke 无模型 10
cv classification mlp_mixer_b16_224 flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=6680d967c8d66d5ad6716cb3e1d4e63a
cv classification mlp_mixer_b32_224 flowvision ke 无模型 10
cv classification mlp_mixer_b16_224_in21k flowvision ke infer低 10
cv classification mlp_mixer_l16_224 flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=9d39de61bbaa96f995494981606131bf
cv classification mlp_mixer_l32_224 flowvision ke 无模型 10
cv classification mlp_mixer_l16_224_in21k flowvision ke infer低 10
cv classification mlp_mixer_b16_224_miil flowvision ke infer低 10
cv classification mlp_mixer_b16_224_miil_in21k flowvision ke infer低 10
cv classification convmixer_768_32_relu flowvision ke 完成 3 li https://www.oneflow.cloud/drill/#/project/public/code?id=ea73a6af66d18116bd17159bd43dfd09
cv classification convmixer_1024_20 flowvision ke 完成 3 li https://www.oneflow.cloud/drill/#/project/public/code?id=d4cae7556245781c1da91665be4a4a2a
cv classification convmixer_1536_20 flowvision ke 完成 3 li https://www.oneflow.cloud/drill/#/project/public/code?id=9ab2b324f3643859826de05b0e0b0b09
cv classification res2net101_26w_4s flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=cd1e229dae70bfe32e33d57798e5b62e
cv classification res2net50_14w_8s flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=ab6afe87aa33c278311d63b649f6b117
cv classification res2net50_26w_4s flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=77a0ad1c8078b75cba5d823bca361c04
cv classification res2net50_26w_6s flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=fa0a61d5d72247d22863c9392b48f89c
cv classification res2net50_26w_8s flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=156564fb425715a833a255b3736f9e9d
cv classification res2net50_48w_2s flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=1ddc5d5f68e2d3a75d9a9c8d9a59c97a
cv classification cait_M48 flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=405fc951ffa80c9dcc3a605d41587de4
cv classification cait_M36 flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=c88925b7c61385982b8b25658467ed7d
cv classification cait_S36 flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=a9165b582fca4851946455e1158fa5d5
cv classification cait_S24 flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=ef302a9ec1241961ad2d9ff21dd28a1f
cv classification cait_S24_224 flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=ef9a11c3daa419e068f913d6c1da49be
cv classification cait_XS24 flowvision ke 完成 6 li https://www.oneflow.cloud/drill/#/project/public/code?id=ec1e9e9df444a8012a699020d7876c3f
cv classification dla34 flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=92865fd08123c84dd002ad934dab9f30
cv classification dla46_c flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=4a23dbc75bd431049450b709f1e2dc82
cv classification dla46x_c flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=48e8f80f732ce1ff49401c239079799b
cv classification dla60x_c flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=72a36495db48f1f3f4830108eed027df
cv classification dla60 flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=0adcb76c0580f05ec53d87b3ed2689bb
cv classification dla60x flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=a44cc9aa071ec6171c0aae3ccc7eea69
cv classification dla102 flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=df2140e6b6ff3c0bebc2a05782bf26ff
cv classification dla102x flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=c134390359f1d99647ffc6e814b4b59a
cv classification dla102x2 flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=a1a5a8c0d0038728372b98116df242db
cv classification dla169 flowvision ke 完成 10 li https://www.oneflow.cloud/drill/#/project/public/code?id=b02b358f69982312ce01028c25325a3a
cv classification genet_small flowvision ke 完成 3 li https://www.oneflow.cloud/drill/#/project/public/code?id=32e02310d717d8fa2c6841cf44b1b398
cv classification genet_normal flowvision ke 完成 3 li https://www.oneflow.cloud/drill/#/project/public/code?id=d0ff57b9dd41286da62eeff48365f393
cv classification genet_large flowvision ke 完成 3 li https://www.oneflow.cloud/drill/#/project/public/code?id=e395406e16ce22f84d5c6c75945da24c
cv classification hrnet_w18_small flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=1fd560ba809b25931f26285ee87fbd01
cv classification hrnet_w18_small_v2 flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=ed3a0a67216e00c5e16ca68297e2e51c
cv classification hrnet_w18 flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=477afc03da45d95475c585087263aa16
cv classification hrnet_w30 flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=b7830e267e11d1c24dc2e1f809815b58
cv classification hrnet_w32 flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=02b78b105d109ee2d73b141a84131d5d
cv classification hrnet_w40 flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=c4ebccc6d6c51c4726268327bf0d1277
cv classification hrnet_w44 flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=4ad93af70e75ad40b35b3a4037b42cd2
cv classification hrnet_w48 flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=5260c81e98fb3e315b211143e9fc3d1f
cv classification hrnet_w64 flowvision ke 完成 9 li https://www.oneflow.cloud/drill/#/project/public/code?id=f16740051ea0350f2c4a7cdbcf3f03c4
cv classification fan_vit_tiny flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=5fdc757c88feab49cf3cd5321e191fcc
cv classification fan_vit_small flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=f5e4fc22ddab8c1bacf86bc06b1b4f3a
cv classification fan_vit_large flowvision zhang 没有模型 12 li
cv classification fan_hybrid_tiny flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=91286420dfed2958fcdaa44947f5fdd6
cv classification fan_hybrid_small flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=0f6c5f287e19e86199d1c651353683bc
cv classification fan_hybrid_base flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=dc39a54b6125b8a025d8ee9aa9a56d7e
cv classification fan_hybrid_large flowvision zhang 没有模型 12 li
cv classification fan_base_18_p16_224 flowvision li 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=f6af8debe2b2cac9ac6a78b6640131fd
cv classification fan_hybrid_base_in22k_1k flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=4e48d6211354973f5a25d684e55e204b
cv classification fan_hybrid_base_in22k_1k_384 flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=727170a01151d1b960dae29560aefd1b
cv classification fan_hybrid_large_in22k_1k flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=8fa9548610a845ebde39fe4a9e36e680
cv classification fan_hybrid_large_in22k_1k_384 flowvision zhang 完成 12 li https://www.oneflow.cloud/drill/#/project/public/code?id=932d7f4b229527e61eeedc46d90f77c3
cv classification pvt_small flowvision ke 完成 4 li https://www.oneflow.cloud/drill/#/project/public/code?id=6f2d36e01ff216a91e884e106cc61b01
cv classification pvt_tiny flowvision zhang 完成 4 li https://www.oneflow.cloud/drill/#/project/public/code?id=68109071b32382961de42e6b2f6d38b1
cv classification pvt_medium flowvision zhang 完成 4 li https://www.oneflow.cloud/drill/#/project/public/code?id=f90ccd277fd99b91728907fbf4ad6052
cv classification pvt_large flowvision zhang 完成 4 li https://www.oneflow.cloud/drill/#/project/public/code?id=6b5533383abbf289ab905a6f902b3bbb
cv classification regionvit_base flowvision ke 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=8a772ef77e8265dad8e4ef6699b52c61
cv classification van_base flowvision ke 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=3e9a2df19f47b4325144885e3a37cbc1
cv classification AlexNet flowvision ke 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=72080813882acf841b8de45e88102f3c 
cv classification SqueezeNet flowvision ke 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=0a5e15b73b1c8289ef43439e3fdac6ba 
cv classification SqueezeNet 1.1 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=46d7e0a2ef7f55aa9cf82b37a7c530c2
cv classification VGG-11 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=61c0fe736758792228d4e0bfb27beb04
cv classification VGG-11-BN flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=381d00d5d6ce5ca40dcedd003d98b94a
cv classification VGG-13 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=9bc8a6c17172722a7e04cf74f8b3b4de
cv classification VGG-13-BN flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=77e3480052ab9bacd2b37caeca5885e8
cv classification VGG-16 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=6166d5d2af2b1bd29edadf88b62ef975
cv classification VGG-16-BN flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=89ac74c572ee6833d0a70d1650e1708a
cv classification VGG-19 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=12256eee1b678bbe45f8610caba00f96
cv classification VGG-19-BN flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=e84034d9edd66977cd70130268dda4e0
cv classification GoogLeNet flowvision zhang 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=0f49a25e9fd85d21c2ae06e200452112
cv classification Inception_V3 flowvision zhang 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=3648dac8652969bdd49848071d0f968a
cv classification ResNet-18 flowvision ke 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=52e1e7d51e9f340ee18101edac59cc6d 
cv classification ResNet-34 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=7ba5a5946ee5bdade3a6ed0d421a28d1
cv classification ResNet-50 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=64669893332daab2ae874dfe839abd88
cv classification ResNet-101 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=d069d9ce05d84290518f1a79c8db0f15
cv classification ResNet-152 flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=e60699fb4e75a3c9a6efe451c3c4dc4c
cv classification ResNeXt-50 32x4d flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=fa231a90a53651b92b9642b56c29dd19
cv classification ResNeXt-101 32x8d flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=0eb0c2b9f04381e99a8ed99b0ecf6e8a
cv classification ResNeSt-50 flowvision zhang 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=3669763ea9382906b0132dfb04c50c47
cv classification ResNeSt-101 flowvision zhang 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=b3b28a3065d162df62df6144b3cc8f6e
cv classification ResNeSt-200 flowvision zhang 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=41f5ecf0e37f60b17fa697b972ab5738
cv classification ResNeSt-269 flowvision zhang 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=9d3d1ab2792a86a9d3c02887a9a3b1d2
cv classification SE-ResNet101 flowvision zhang 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=6b50454e5ef09a31c40ada924c199cf9
cv classification SE-ResNet152 flowvision zhang 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=00ffee25af16d5951ec01e6fcaba5a29
cv classification SE-ResNet50 flowvision zhang 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=87aec92114ebc62e5b29575454a29f3f
cv classification SE-ResNeXt101-32x4d flowvision zhang 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=763356489039b10c91a635f3f59e4029
cv classification SE-ResNeXt50-32x4d flowvision zhang 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=21fefb3568303b5d0328e64b76bf1bfc
cv classification SENet-154 flowvision zhang 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=eb114bd5bc28f464ee624db70abb1a30
cv classification DenseNet-121 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=9383d1ee8cf6b0beb1d82e001327e7e6
cv classification DenseNet-161 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=748e5cf1d4d07c3439e61102f41503d1
cv classification DenseNet-169 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=2c31515668a6c3d01ac9dbe5d35f4051
cv classification DenseNet-201 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=0ed8889b7be3e26a4e0b821cc584380d
cv classification ShuffleNet_V2 x0.5 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=75a9ebdce52dbaea88ea921df32a36f2
cv classification ShuffleNet_V2 x1.0 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=90d9a3ddcb8bafabc951742ebc4defd5
cv classification ShuffleNet_V2 x1.5 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=569fe5043eda185002ab35a08d9707d7
cv classification ShuffleNet_V2 x2.0 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=50d954efb04235e6b9f993cbb9492457
cv classification MobileNet_V2 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=8fac67899e5bdcc0981da13c3cab1b89
cv classification MobileNet_V3 small flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=e5fd35642b7ac9b537110ceb70c9832a
cv classification MobileNet_V3 large flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=6529be091cb9c1af73fbe43f85092dee
cv classification MNASNet x0.5 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=9c2419d250dc5b2d61c47fdbcbc45b23
cv classification MNASNet x0.75 flowvision cui 完成    li https://www.oneflow.cloud/drill/#/project/public/code?id=dea56da93c65719360b637a842096c7e 
cv classification MNASNet x1.0 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=92dd8f01260cf7a01db8bc57f0d44cf9
cv classification MNASNet x1.3 flowvision cui 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=f09034e44f7fa3653a108a486defe1a9
cv classification GhostNet flowvision ke 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=8b02840745f682ae928da9bc618e6bb0
cv classification CrossFormer-T flowvision zhang 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=ec876cce29a7d9da57f3f01bcac57bb8
cv classification CrossFormer-S flowvision zhang 完成 li  https://www.oneflow.cloud/drill/#/project/public/code?id=31ab1b82aa32df9fd9eca0d99226c6a2
cv classification CrossFormer-B flowvision zhang 完成 li  https://www.oneflow.cloud/drill/#/project/public/code?id=ee4ecc63d59f0e26e178185adf5b70e4
cv classification CrossFormer-L flowvision zhang 完成 li  https://www.oneflow.cloud/drill/#/project/public/code?id=bc3f08aa3513464d10aab15a75a67fa5
cv classification PoolFormer-S12 flowvision zhou 完成  li https://www.oneflow.cloud/drill/#/project/public/code?id=3802898d0df4ab05479425c56cbdc5d8
cv classification PoolFormer-S24 flowvision zhang 完成  li  https://www.oneflow.cloud/drill/#/project/public/code?id=08b8873535d3179796697808cd470bba
cv classification PoolFormer-S36 flowvision zhang 完成  li https://www.oneflow.cloud/drill/#/project/public/code?id=e76ce4d08ffc55277ec18b584acf0d07 
cv classification PoolFormer-M36 flowvision zhang 完成  li  https://www.oneflow.cloud/drill/#/project/public/code?id=03654a36457379ae02da2cda230cf8ef
cv classification PoolFormer-M48 flowvision zhang 完成  li  https://www.oneflow.cloud/drill/#/project/public/code?id=52fe605fc822974a72a359c5ca68a8e2
cv classification gMLP flowvision ke 完成 li https://www.oneflow.cloud/drill/#/project/public/code?id=db86fa73ae42ec1138bbf9c282d715a7
cv classification ConvNeXt flowvision ke 完成 可拓展至：18 li https://www.oneflow.cloud/drill/#/project/public/code?id=580631c6005d96e2fc756b697264911c
cv classification LeViT flowvision ke 完成(infer低) 可拓展至:5 li https://www.oneflow.cloud/drill/#/project/public/code?id=e53b9e1e52602aa6887e8289aed2ae42
cv classification MobileViT flowvision li 进行中（infer低）
cv Semantic Segmentation fcn_resnet101_coco flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=5a67c4d8f94d50904b00e1ab7dc073b0
cv Semantic Segmentation fcn_resnet50_coco flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=5b958d4676e6a873a85f50f7327d5810
cv Semantic Segmentation deeplabv3_mobilenet_v3_large_coco flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=7fcd75aeda576e41fcc8841a5d38c44d
cv Semantic Segmentation deeplabv3_resnet101_coco flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=98a1dbe2c439f5fae5c24ba7c6109c47
cv Semantic Segmentation deeplabv3_resnet50_coco flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=56ee96d906df67c2af2390d36dffb694
cv Semantic Segmentation lraspp_mobilenet_v3_large_coco flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=0836620e36d9c728547fde5c08d2939d
cv Object Detection fasterrcnn_mobilenet_v3_large_320_fpn flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=83955d0dbd8f6e398384297300031d40
cv Object Detection fasterrcnn_mobilenet_v3_large_fpn flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=123c9a849cc8180206e8cfe42e32c7e3
cv Object Detection fasterrcnn_resnet50_fpn flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=3e9d4f6d8d81dd3626ff3a2491a4ece3
cv Object Detection maskrcnn_resnet50_fpn flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=bd90c25e18860ac4c4f669e0a7a2dadc
cv Object Detection retinanet_resnet50_fpn flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=965abe47aeac05155363492ec6b6d6c7
cv Object Detection ssd300_vgg16 flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=c677f6ab3885b5be9fd482575ee25a21
cv Object Detection ssdlite320_mobilenet_v3_large flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=aacf292f065ecfbeee4b3df54bde6e55
cv Object Detection fcos_resnet50_fpn flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=dd9230b9351651ffce3df71878fcb470
cv Neural Style Transfer style_transfer.fast_neural_style flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=0006f425af1fee25d4dc14385a38a4e6
cv Face Recognition iresnet50 flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=a61c970743904dc4773e40b0abecd064
cv Face Recognition iresnet101 flowvision zhou 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=b4804fb168d41aa3a94f83abd0019a7d
cv   VisionTransformer libai li 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=dc07b87d3a60e36d182f908e8bb660f9
nlp   SwinTransformer libai li 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=62f0d66aa7441b9f10f877176af27e7d
nlp   SwinTransformerV2 libai li 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=47260dd33f92bedfc1f024095dcbc035
nlp   ResMLP libai li 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=f76e38f8228c97ecadfe0482ed2fc06c
nlp   BERT libai li 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=c972180739924ca78d89f664b7ad135f
nlp   RoBERTa libai li 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=d3bbb391e714b1f780aac86e3aa074a9
nlp   T5 libai li 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=6152fadc8e6997f972ce1a85841daf78
nlp   GPT-2 libai li 完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=161fb0c7e6c609526b97bd30d935e1b8
nlp text_classfication Transformer CoModels maolin  完成   li https://www.oneflow.cloud/drill/#/project/public/code?id=1ad00d3d7db8c504b08345461b3999b0
nlp odd_numbers Transformer CoModels maolin 完成    li https://www.oneflow.cloud/drill/#/project/public/code?id=c6bd57e6b4f6362348783da981f6f0ff
science Equation inversion-Lorenz system PINNs CoModels zhang 完成    li https://www.oneflow.cloud/drill/#/project/public/code?id=91f8fb23498318f83dd16e0e875db6bc
science Fluid simulation-ldc PINNs CoModels zhang  完成    li https://www.oneflow.cloud/drill/#/project/public/code?id=b1914ead9370947a60b3a9c3ee60b0c0
akeeei commented 11 months ago

vit_tiny_patch16_224

cd CoModels/cv/classification/vit_tiny_patch16_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 32
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 0.003
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_tiny_patch16_224 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/92c39461-2f9d-445b-9533-e2bda5cea595) * 训练结果 ： ``` INFO * Acc@1 53.941 Acc@5 78.964 INFO Accuracy of the network on the 98 test images: 53.9% INFO Max accuracy: 64.19% INFO Training time 6:49:33 ```
推理结果 ``` INFO * Acc@1 45.267 Acc@5 68.073 INFO Accuracy of the network on the 1563 test images: 45.3% INFO throughput averaged with 30 times INFO batch_size 32 throughput 1757.057121686971 ```
akeeei commented 11 months ago

vit_tiny_patch16_384

cd CoModels/cv/classification/vit_tiny_patch16_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 32
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 384

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_tiny_patch16_384 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/23aefd6f-f99c-44bb-a418-e40240c5ddb7) * 训练结果 ： ``` INFO * Acc@1 68.616 Acc@5 89.426 INFO Accuracy of the network on the 98 test images: 68.6% INFO Max accuracy: 68.62% INFO Training time 1:40:07 ```
推理结果 ``` INFO * Acc@1 48.120 Acc@5 70.708 INFO Accuracy of the network on the 1563 test images: 48.1% INFO throughput averaged with 30 times INFO batch_size 32 throughput 328.12604634424923 ```
akeeei commented 11 months ago

vit_small_patch32_224

cd CoModels/cv/classification/vit_small_patch32_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 256
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_small_patch32_224 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/0ccf4fec-c50c-42d4-9584-bd387b7070c6) ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/732823e2-ef47-49b4-abab-280d18ec357f) * 训练结果 ： ``` INFO * Acc@1 66.895 Acc@5 87.109 INFO Accuracy of the network on the 49 test images: 66.9% INFO Max accuracy: 67.58% INFO Training time 0:57:20 ```
推理结果 ``` INFO * Acc@1 62.294 Acc@5 84.081 INFO Accuracy of the network on the 1563 test images: 62.3% INFO throughput averaged with 30 times INFO batch_size 32 throughput 2159.266911487113 ```
akeeei commented 11 months ago

vit_small_patch32_384

cd CoModels/cv/classification/vit_small_patch32_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 128
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 384

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_small_patch32_384 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/a49cb950-8d6d-4079-bb4e-901d7d82be92) ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/33ea5b57-311e-47c1-8155-c6594022953b) * 训练结果 ： ``` INFO * Acc@1 73.145 Acc@5 93.945 INFO Accuracy of the network on the 49 test images: 73.1% INFO Max accuracy: 74.12% INFO Training time 2:22:48 ```
推理结果 ``` INFO * Acc@1 68.424 Acc@5 88.513 INFO Accuracy of the network on the 1563 test images: 68.4% INFO throughput averaged with 30 times INFO batch_size 32 throughput 965.2715564422288 ```
akeeei commented 11 months ago

vit_small_patch16_384

cd CoModels/cv/classification/vit_small_patch16_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 32
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 0.003
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_small_patch16_384 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/ecc204b8-b60e-4e47-b5e7-7bf34ecffc1e) ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/ca1cb0c2-ed08-4bdb-9869-59eaf52a99eb) * 训练结果 ： ``` INFO * Acc@1 76.015 Acc@5 93.935 INFO Accuracy of the network on the 391 test images: 76.0% INFO Max accuracy: 76.01% INFO Training time 1:22:44 ```
推理结果 ``` INFO * Acc@1 76.932 Acc@5 93.956 INFO Accuracy of the network on the 1563 test images: 76.9% INFO throughput averaged with 30 times INFO batch_size 32 throughput 425.0662919764819 ```
akeeei commented 11 months ago

vit_base_patch32_224

cd CoModels/cv/classification/vit_base_patch32_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 256
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_base_patch32_224 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/ca3e3cad-5aae-49fa-a9d9-136e0fa2a782) ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/83c8a3f0-5505-4a93-be0b-750b070b708e) * 训练结果 ： ``` INFO * Acc@1 77.516 Acc@5 93.892 INFO Accuracy of the network on the 98 test images: 77.5% INFO Max accuracy: 77.52% INFO Training time 0:41:03 ```
推理结果 ``` INFO * Acc@1 72.105 Acc@5 90.895 INFO Accuracy of the network on the 1563 test images: 72.1% INFO throughput averaged with 30 times INFO batch_size 32 throughput 880.851543616409 ```
akeeei commented 11 months ago

vit_base_patch32_384

cd CoModels/cv/classification/vit_base_patch32_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 128
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 384

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_base_patch32_384 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/951658c5-645b-4273-9a40-68769903551f) * 训练结果 ： ``` INFO * Acc@1 74.982 Acc@5 93.119 INFO Accuracy of the network on the 196 test images: 75.0% INFO Max accuracy: 74.98% INFO Training time 1:21:08 ```
推理结果 ``` INFO * Acc@1 76.455 Acc@5 93.238 INFO Accuracy of the network on the 1563 test images: 76.5% INFO throughput averaged with 30 times INFO batch_size 32 throughput 341.92431427372634 ```
akeeei commented 11 months ago

vit_base_patch16_384

cd CoModels/cv/classification/vit_base_patch16_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 32
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 384

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_base_patch16_384 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/d95d5c70-6cf8-458a-9bd2-e4bbb05faf20) * 训练结果 ： ``` INFO * Acc@1 73.707 Acc@5 92.468 INFO Accuracy of the network on the 391 test images: 73.7% INFO Max accuracy: 73.71% INFO Training time 2:21:20 ```
推理结果 ``` INFO * Acc@1 78.719 Acc@5 94.443 INFO Accuracy of the network on the 1563 test images: 78.7% INFO throughput averaged with 30 times INFO batch_size 32 throughput 60.235781614383676 ```
akeeei commented 11 months ago

vit_base_patch8_224

cd CoModels/cv/classification/vit_base_patch8_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 16
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 3e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_base_patch8_224 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/6eb2c47b-217a-441e-ab5f-7f249cdcca4b) * 训练结果 ： ``` INFO * Acc@1 80.505 Acc@5 95.880 INFO Accuracy of the network on the 782 test images: 80.5% INFO Max accuracy: 80.51% INFO Training time 4:41:53 ```
推理结果 ``` INFO * Acc@1 76.249 Acc@5 92.587 INFO Accuracy of the network on the 1563 test images: 76.2% INFO throughput averaged with 30 times INFO batch_size 32 throughput 35.069299488421656 ```
kokuro-asahi commented 11 months ago

EfficientNet_b0

Training ``` cd CoModels/cv/classification/efficientnet_b0 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 32 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: efficientnet_b0 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/efficientnet_b0/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.1 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 1.25e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/efficientnet_b0 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/239cb36f-456c-47fa-9a94-e36eac35442e)
推理结果 INFO * Acc@1 77.704 Acc@5 93.547
kokuro-asahi commented 11 months ago

EfficientNet_b1

Training ``` cd CoModels/cv/classification/efficientnet_b1 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 32 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: efficientnet_b0 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/efficientnet_b1/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.1 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 1.25e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/efficientnet_b1 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/38f52319-4381-423b-bc37-2203c1ffc344)
推理结果 INFO * Acc@1 77.589 Acc@5 93.641 NFO Accuracy of the network on the 1563 test images: 77.6%
kokuro-asahi commented 11 months ago

EfficientNet_b2

Training ``` cd CoModels/cv/classification/efficientnet_b2 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 32 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: efficientnet_b2 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/efficientnet_b2/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.1 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 1.25e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/efficientnet_b2 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/57fa0b57-98ac-45fb-8c39-85ef00f0b839)
推理结果 INFO * Acc@1 57.904 Acc@5 78.367
akeeei commented 11 months ago

vit_small_patch16_224

cd CoModels/cv/classification/vit_small_patch16_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 256
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 3e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_small_patch16_224 bash infer.sh ```
训练过程 * 训练日志 ： ![281599663-a6910294-5e73-467b-9b9b-b1485b62c45e](https://github.com/Oneflow-Inc/CoModels/assets/98879022/55e72e3f-a64b-4cf2-9feb-75a106fb308a) * 训练结果 ： ``` INFO * Acc@1 71.530 Acc@5 91.356 INFO Accuracy of the network on the 98 test images: 71.5% INFO Max accuracy: 71.53% INFO Training time 0:33:14 ```
推理结果 ``` INFO * Acc@1 74.206 Acc@5 92.386 INFO Accuracy of the network on the 1563 test images: 74.2% INFO throughput averaged with 30 times INFO batch_size 32 throughput 725.5174421915012 ```
akeeei commented 11 months ago

vit_large_patch32_384

cd CoModels/cv/classification/vit_large_patch32_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 64
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 384

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 3e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_large_patch32_384 bash infer.sh ```
训练过程 * 训练日志 ： ![图片](https://github.com/Oneflow-Inc/CoModels/assets/98879022/31b64638-33c4-48bc-802a-067bbf488f86) ![图片](https://github.com/Oneflow-Inc/CoModels/assets/98879022/1f09f44a-0677-4f1c-9293-53b9b340c16f) * 训练结果 ： ``` INFO * Acc@1 79.412 Acc@5 95.099 INFO Accuracy of the network on the 391 test images: 79.4% INFO Max accuracy: 79.41% INFO Training time 2:27:31 ```
推理结果 ``` INFO * Acc@1 75.623 Acc@5 93.137 INFO Accuracy of the network on the 1563 test images: 75.6% INFO throughput averaged with 30 times INFO batch_size 32 throughput 245.38814350421615 ```
akeeei commented 11 months ago

vit_large_patch16_224

cd CoModels/cv/classification/vit_large_patch16_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 32
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_large_patch16_224 bash infer.sh ```
训练过程 * 训练日志 ： ![图片](https://github.com/Oneflow-Inc/CoModels/assets/98879022/fff957f8-9f3a-4ba5-9619-afc2d25ab494) * 训练结果 ： ``` INFO * Acc@1 84.833 Acc@5 97.480 INFO Accuracy of the network on the 782 test images: 84.8% INFO Max accuracy: 84.83% INFO Training time 3:28:39 ```
推理结果 ``` INFO * Acc@1 83.462 Acc@5 96.732 INFO Accuracy of the network on the 1563 test images: 83.5% INFO throughput averaged with 30 times INFO batch_size 32 throughput 184.42577691807244 ```
akeeei commented 11 months ago

vit_base_patch16_224_sam

cd CoModels/cv/classification/vit_base_patch16_224_sam
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 128
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_base_patch16_224_sam bash infer.sh ```
训练过程 * 训练日志 ： ![图片](https://github.com/Oneflow-Inc/CoModels/assets/98879022/c4ceab03-fc10-4a23-be9b-a7e415fb177e) * 训练结果 ： ``` INFO * Acc@1 78.311 Acc@5 93.732 INFO Accuracy of the network on the 196 test images: 78.3% INFO Max accuracy: 78.31% INFO Training time 0:57:02 ```
推理结果 ``` INFO * Acc@1 75.530 Acc@5 92.172 INFO Accuracy of the network on the 1563 test images: 75.5% INFO throughput averaged with 30 times INFO batch_size 32 throughput 267.0716634529662 ```
akeeei commented 11 months ago

vit_base_patch32_224_sam

cd CoModels/cv/classification/vit_base_patch32_224_sam
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 128
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_base_patch32_224_sam bash infer.sh ```
训练过程 * 训练日志 ： ![图片](https://github.com/Oneflow-Inc/CoModels/assets/98879022/6589374a-bf9d-4c6e-b8c8-2bdc5fa3b89b) * 训练结果 ： ``` INFO * Acc@1 70.874 Acc@5 89.295 INFO Accuracy of the network on the 196 test images: 70.9% INFO Max accuracy: 70.87% INFO Training time 2:21:20 ```
推理结果 ``` INFO * Acc@1 63.574 Acc@5 83.926 INFO Accuracy of the network on the 1563 test images: 63.6% INFO throughput averaged with 30 times INFO batch_size 32 throughput 1071.7754732650496 ```
akeeei commented 11 months ago

vit_large_patch16_384

cd CoModels/cv/classification/vit_large_patch16_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 8
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 384

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 300
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: cosine

  OPTIMIZER:
    NAME: adamw
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/vit_large_patch16_384 bash infer.sh ```
训练过程 * 训练日志 ： ![图片](https://github.com/Oneflow-Inc/CoModels/assets/98879022/53c84701-bf1d-43e7-92c9-30ce06ba9f85) * 训练结果 ： ``` INFO * Acc@1 85.411 Acc@5 97.703 INFO Accuracy of the network on the 1563 test images: 85.4% INFO Max accuracy: 85.41% INFO Training time 6:10:56 ```
推理结果 ``` INFO * Acc@1 85.094 Acc@5 97.570 INFO Accuracy of the network on the 1563 test images: 85.1% INFO throughput averaged with 30 times INFO batch_size 32 throughput 54.038784254131365 ```
kokuro-asahi commented 11 months ago

EfficientNet_b3

Training ``` cd CoModels/cv/classification/efficientnet_b3 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 32 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: efficientnet_b3 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/efficientnet_b3/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.1 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 1.25e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/efficientnet_b3 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/676a1fcb-e8a2-4fef-82ab-c6c2c6884257)
推理结果 INFO * Acc@1 64.891 Acc@5 74.763
kokuro-asahi commented 11 months ago

EfficientNet_b4

Training ``` cd CoModels/cv/classification/efficientnet_b4 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 32 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: efficientnet_b4 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/efficientnet_b4/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.1 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 1.25e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/efficientnet_b4 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/366df3ab-80ef-4ec3-9893-868be2de2d97)
推理结果 INFO * Acc@1 57.009 Acc@5 62.790
kokuro-asahi commented 11 months ago

EfficientNet_b5

Training ``` cd CoModels/cv/classification/efficientnet_b5 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 32 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: efficientnet_b4 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/efficientnet_b4/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.1 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 1.25e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/efficientnet_b4 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/366df3ab-80ef-4ec3-9893-868be2de2d97)
推理结果 INFO * Acc@1 57.009 Acc@5 62.790
kokuro-asahi commented 11 months ago

EfficientNet_b6

cd CoModels/cv/classification/efficientnet_b6
bash train.sh
训练所用超参数 
AMP_OPT_LEVEL: ''
AUG:
  AUTO_AUGMENT: rand-m9-mstd0.5-inc1
  COLOR_JITTER: 0.4
  CUTMIX: 0.0
  CUTMIX_MINMAX: null
  MIXUP: 0.0
  MIXUP_MODE: batch
  MIXUP_PROB: 1.0
  MIXUP_SWITCH_PROB: 0.5
  RECOUNT: 1
  REMODE: pixel
  REPROB: 0.25
BASE:
- ''
DATA:
  BATCH_SIZE: 32
  CACHE_MODE: part
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224
  INTERPOLATION: bicubic
  NUM_CLASSES: 1000
  NUM_WORKERS: 4
  PIN_MEMORY: true
  SYNTHETIC_DATA: false
  ZIP_MODE: false
EVAL_MODE: true
LOCAL_RANK: 0
MODEL:
  ARCH: efficientnet_b6
  CHECKPOINTS: null
  DROP_PATH_RATE: 0.1
  DROP_RATE: 0.0
  LABEL_SMOOTHING: 0.1
  NUM_CLASSES: 1000
  PRETRAINED: true
  RESUME: ''
OUTPUT: output/efficientnet_b6/default
PRINT_FREQ: 50
SAVE_FREQ: 1
SEED: 42
TAG: default
TEST:
  CROP: true
  SEQUENTIAL: false
THROUGHPUT_MODE: false
TRAIN:
  ACCUMULATION_STEPS: 0
  AUTO_RESUME: false
  BASE_LR: 0.1
  CLIP_GRAD: 5.0
  EPOCHS: 90
  LR_SCHEDULER:
    DECAY_EPOCHS: 30
    DECAY_RATE: 0.1
    MILESTONES:
    - 150
    - 225
    NAME: step
  MIN_LR: 1.25e-06
  OPTIMIZER:
    BETAS:
    - 0.9
    - 0.999
    EPS: 1.0e-08
    MOMENTUM: 0.9
    NAME: sgd
  START_EPOCH: 0
  USE_CHECKPOINT: false
  WARMUP_EPOCHS: 0
  WARMUP_LR: 5.0e-07
  WEIGHT_DECAY: 0.0001

Inference ``` cd CoModels/cv/classification/efficientnet_b6 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/4439bb0b-ef41-4ad2-baa0-02bd1a766d9f)
推理结果 INFO * Acc@1 47.688 Acc@5 58.346
kokuro-asahi commented 11 months ago

EfficientNet_b7

cd CoModels/cv/classification/efficientnet_b7
bash train.sh
训练所用超参数 
AMP_OPT_LEVEL: ''
AUG:
  AUTO_AUGMENT: rand-m9-mstd0.5-inc1
  COLOR_JITTER: 0.4
  CUTMIX: 0.0
  CUTMIX_MINMAX: null
  MIXUP: 0.0
  MIXUP_MODE: batch
  MIXUP_PROB: 1.0
  MIXUP_SWITCH_PROB: 0.5
  RECOUNT: 1
  REMODE: pixel
  REPROB: 0.25
BASE:
- ''
DATA:
  BATCH_SIZE: 32
  CACHE_MODE: part
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224
  INTERPOLATION: bicubic
  NUM_CLASSES: 1000
  NUM_WORKERS: 4
  PIN_MEMORY: true
  SYNTHETIC_DATA: false
  ZIP_MODE: false
EVAL_MODE: true
LOCAL_RANK: 0
MODEL:
  ARCH: efficientnet_b7
  CHECKPOINTS: null
  DROP_PATH_RATE: 0.1
  DROP_RATE: 0.0
  LABEL_SMOOTHING: 0.1
  NUM_CLASSES: 1000
  PRETRAINED: true
  RESUME: ''
OUTPUT: output/efficientnet_b7/default
PRINT_FREQ: 50
SAVE_FREQ: 1
SEED: 42
TAG: default
TEST:
  CROP: true
  SEQUENTIAL: false
THROUGHPUT_MODE: false
TRAIN:
  ACCUMULATION_STEPS: 0
  AUTO_RESUME: false
  BASE_LR: 0.1
  CLIP_GRAD: 5.0
  EPOCHS: 90
  LR_SCHEDULER:
    DECAY_EPOCHS: 30
    DECAY_RATE: 0.1
    MILESTONES:
    - 150
    - 225
    NAME: step
  MIN_LR: 1.25e-06
  OPTIMIZER:
    BETAS:
    - 0.9
    - 0.999
    EPS: 1.0e-08
    MOMENTUM: 0.9
    NAME: sgd
  START_EPOCH: 0
  USE_CHECKPOINT: false
  WARMUP_EPOCHS: 0
  WARMUP_LR: 5.0e-07
  WEIGHT_DECAY: 0.0001

Inference ``` cd CoModels/cv/classification/efficientnet_b7 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/b567662e-79bb-45f2-a3c6-75130f411bcc)
推理结果 INFO * Acc@1 59.689 Acc@5 83.967
akeeei commented 11 months ago

deit_tiny_patch16_224

cd CoModels/cv/classification/deit_tiny_patch16_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 256
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 5e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/deit_tiny_patch16_224 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/45036f7e-cdc3-474a-96aa-f3459ab1590f) * 训练结果 ： ``` INFO * Acc@1 72.266 Acc@5 91.309 INFO Accuracy of the network on the 49 test images: 72.3% INFO Max accuracy: 72.27% INFO Training time 0:38:55 ```
推理结果 ``` INFO * Acc@1 72.141 Acc@5 91.159 INFO Accuracy of the network on the 1563 test images: 72.1% INFO throughput averaged with 30 times INFO batch_size 32 throughput 1955.4064216855359 ```
akeeei commented 11 months ago

deit_base_patch16_224

cd CoModels/cv/classification/deit_base_patch16_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 32
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 5e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/deit_base_patch16_224 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/5eb2b55e-9279-4d26-8b14-db78ee841339) * 训练结果 ： ``` INFO * Acc@1 81.535 Acc@5 95.444 INFO Accuracy of the network on the 391 test images: 81.5% INFO Max accuracy: 81.53% INFO Training time 2:39:41 ```
推理结果 ``` INFO * Acc@1 81.816 Acc@5 95.592 INFO Accuracy of the network on the 1563 test images: 81.8% INFO throughput averaged with 30 times INFO batch_size 32 throughput 134.9496680093491 ```
akeeei commented 11 months ago

deit_base_patch16_384

cd CoModels/cv/classification/deit_base_patch16_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 16
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 384

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/deit_base_patch16_384 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/83e56bbd-3c22-4ae8-8456-cd7e6133fa61) * 训练结果 ： ``` INFO * Acc@1 82.885 Acc@5 96.207 INFO Accuracy of the network on the 782 test images: 82.9% INFO Max accuracy: 82.96% INFO Training time 6:03:41 ```
推理结果 ``` INFO * Acc@1 82.872 Acc@5 96.234 INFO Accuracy of the network on the 1563 test images: 82.9% INFO throughput averaged with 30 times INFO batch_size 32 throughput 52.672114994542234 ```
kokuro-asahi commented 11 months ago

regnet_y_400mf

Training ``` cd CoModels/cv/classification/regnet_y_400mf bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 128 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: regnet_y_400mf CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/regnet_y_400mf/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.4 CLIP_GRAD: 5.0 EPOCHS: 100 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: cosine MIN_LR: 5.0e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 5 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 5.0e-05 ```
Inference ``` cd CoModels/cv/classification/regnet_y_400mf bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/6ba96a14-79ac-43fc-9956-fab97d4de4eb)
推理结果 INFO * Acc@1 78.245 Acc@5 93.542
kokuro-asahi commented 11 months ago

regnet_y_800mf

Training ``` cd CoModels/cv/classification/regnet_y_800mf bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 128 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: regnet_y_800mf CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/regnet_y_800mf/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.4 CLIP_GRAD: 5.0 EPOCHS: 100 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: cosine MIN_LR: 5.0e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 5 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 5.0e-05 ```
Inference ``` cd CoModels/cv/classification/regnet_y_800mf bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/a9d048df-8c4d-4328-9b78-ed0025521bba)
推理结果 INFO * Acc@1 58.245 Acc@5 72.457
kokuro-asahi commented 11 months ago

regnet_y_1_6gf

Training ``` cd CoModels/cv/classification/regnet_y_1_6gf bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 128 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: regnet_y_1_6gf CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/regnet_y_1_6gf/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.4 CLIP_GRAD: 5.0 EPOCHS: 100 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: cosine MIN_LR: 5.0e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 5 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 5.0e-05 ```
Inference ``` cd CoModels/cv/classification/regnet_y_1_6gf bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/b3356cf1-0c7b-47c7-9906-03b6327b4cf9)
推理结果 INFO * Acc@1 40.984 Acc@5 69.236
kokuro-asahi commented 11 months ago

regnet_y_3_2gf

Training ``` cd CoModels/cv/classification/regnet_y_3_2gf bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 128 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: regnet_y_3_2gf CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/regnet_y_3_2gf/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.4 CLIP_GRAD: 5.0 EPOCHS: 100 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: cosine MIN_LR: 5.0e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 5 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 5.0e-05 ```
Inference ``` cd CoModels/cv/classification/regnet_y_3_2gf bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/41c93e42-d0ce-4a05-879d-eae5c15c4386)
推理结果 INFO * Acc@1 62.896 Acc@5 78.919
kokuro-asahi commented 11 months ago

regnet_y_8gf

cd CoModels/cv/classification/regnet_y_8gf
bash train.sh
训练所用超参数 
AMP_OPT_LEVEL: ''
AUG:
  AUTO_AUGMENT: rand-m9-mstd0.5-inc1
  COLOR_JITTER: 0.4
  CUTMIX: 0.0
  CUTMIX_MINMAX: null
  MIXUP: 0.0
  MIXUP_MODE: batch
  MIXUP_PROB: 1.0
  MIXUP_SWITCH_PROB: 0.5
  RECOUNT: 1
  REMODE: pixel
  REPROB: 0.25
BASE:
- ''
DATA:
  BATCH_SIZE: 128
  CACHE_MODE: part
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224
  INTERPOLATION: bicubic
  NUM_CLASSES: 1000
  NUM_WORKERS: 4
  PIN_MEMORY: true
  SYNTHETIC_DATA: false
  ZIP_MODE: false
EVAL_MODE: true
LOCAL_RANK: 0
MODEL:
  ARCH: regnet_y_8gf
  CHECKPOINTS: null
  DROP_PATH_RATE: 0.1
  DROP_RATE: 0.0
  LABEL_SMOOTHING: 0.1
  NUM_CLASSES: 1000
  PRETRAINED: true
  RESUME: ''
OUTPUT: output/regnet_y_8gf/default
PRINT_FREQ: 50
SAVE_FREQ: 1
SEED: 42
TAG: default
TEST:
  CROP: true
  SEQUENTIAL: false
THROUGHPUT_MODE: false
TRAIN:
  ACCUMULATION_STEPS: 0
  AUTO_RESUME: false
  BASE_LR: 0.4
  CLIP_GRAD: 5.0
  EPOCHS: 100
  LR_SCHEDULER:
    DECAY_EPOCHS: 30
    DECAY_RATE: 0.1
    MILESTONES:
    - 150
    - 225
    NAME: cosine
  MIN_LR: 5.0e-06
  OPTIMIZER:
    BETAS:
    - 0.9
    - 0.999
    EPS: 1.0e-08
    MOMENTUM: 0.9
    NAME: sgd
  START_EPOCH: 0
  USE_CHECKPOINT: false
  WARMUP_EPOCHS: 5
  WARMUP_LR: 5.0e-07
  WEIGHT_DECAY: 5.0e-05

Inference ``` cd CoModels/cv/classification/regnet_y_8gf bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/6d6f7975-309d-464d-b164-bd3bc2f1f54f)
推理结果 INFO * Acc@1 68.257 Acc@5 84.568
kokuro-asahi commented 11 months ago

regnet_y_16gf

cd CoModels/cv/classification/regnet_y_16gf
bash train.sh
训练所用超参数 
AMP_OPT_LEVEL: ''
AUG:
  AUTO_AUGMENT: rand-m9-mstd0.5-inc1
  COLOR_JITTER: 0.4
  CUTMIX: 0.0
  CUTMIX_MINMAX: null
  MIXUP: 0.0
  MIXUP_MODE: batch
  MIXUP_PROB: 1.0
  MIXUP_SWITCH_PROB: 0.5
  RECOUNT: 1
  REMODE: pixel
  REPROB: 0.25
BASE:
- ''
DATA:
  BATCH_SIZE: 128
  CACHE_MODE: part
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224
  INTERPOLATION: bicubic
  NUM_CLASSES: 1000
  NUM_WORKERS: 4
  PIN_MEMORY: true
  SYNTHETIC_DATA: false
  ZIP_MODE: false
EVAL_MODE: true
LOCAL_RANK: 0
MODEL:
  ARCH: regnet_y_16gf
  CHECKPOINTS: null
  DROP_PATH_RATE: 0.1
  DROP_RATE: 0.0
  LABEL_SMOOTHING: 0.1
  NUM_CLASSES: 1000
  PRETRAINED: true
  RESUME: ''
OUTPUT: output/regnet_y_16gf/default
PRINT_FREQ: 50
SAVE_FREQ: 1
SEED: 42
TAG: default
TEST:
  CROP: true
  SEQUENTIAL: false
THROUGHPUT_MODE: false
TRAIN:
  ACCUMULATION_STEPS: 0
  AUTO_RESUME: false
  BASE_LR: 0.4
  CLIP_GRAD: 5.0
  EPOCHS: 100
  LR_SCHEDULER:
    DECAY_EPOCHS: 30
    DECAY_RATE: 0.1
    MILESTONES:
    - 150
    - 225
    NAME: cosine
  MIN_LR: 5.0e-06
  OPTIMIZER:
    BETAS:
    - 0.9
    - 0.999
    EPS: 1.0e-08
    MOMENTUM: 0.9
    NAME: sgd
  START_EPOCH: 0
  USE_CHECKPOINT: false
  WARMUP_EPOCHS: 5
  WARMUP_LR: 5.0e-07
  WEIGHT_DECAY: 5.0e-05

Inference ``` cd CoModels/cv/classification/regnet_y_16gf bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/3f4a93ab-1489-40c8-b692-37c9debd4166)
推理结果 INFO * Acc@1 64.886Acc@5 74.743
kokuro-asahi commented 11 months ago

regnet_y_32gf

cd CoModels/cv/classification/regnet_y_32gf
bash train.sh
训练所用超参数 
AMP_OPT_LEVEL: ''
AUG:
  AUTO_AUGMENT: rand-m9-mstd0.5-inc1
  COLOR_JITTER: 0.4
  CUTMIX: 0.0
  CUTMIX_MINMAX: null
  MIXUP: 0.0
  MIXUP_MODE: batch
  MIXUP_PROB: 1.0
  MIXUP_SWITCH_PROB: 0.5
  RECOUNT: 1
  REMODE: pixel
  REPROB: 0.25
BASE:
- ''
DATA:
  BATCH_SIZE: 128
  CACHE_MODE: part
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224
  INTERPOLATION: bicubic
  NUM_CLASSES: 1000
  NUM_WORKERS: 4
  PIN_MEMORY: true
  SYNTHETIC_DATA: false
  ZIP_MODE: false
EVAL_MODE: true
LOCAL_RANK: 0
MODEL:
  ARCH: regnet_y_16gf
  CHECKPOINTS: null
  DROP_PATH_RATE: 0.1
  DROP_RATE: 0.0
  LABEL_SMOOTHING: 0.1
  NUM_CLASSES: 1000
  PRETRAINED: true
  RESUME: ''
OUTPUT: output/regnet_y_32gf/default
PRINT_FREQ: 50
SAVE_FREQ: 1
SEED: 42
TAG: default
TEST:
  CROP: true
  SEQUENTIAL: false
THROUGHPUT_MODE: false
TRAIN:
  ACCUMULATION_STEPS: 0
  AUTO_RESUME: false
  BASE_LR: 0.4
  CLIP_GRAD: 5.0
  EPOCHS: 100
  LR_SCHEDULER:
    DECAY_EPOCHS: 30
    DECAY_RATE: 0.1
    MILESTONES:
    - 150
    - 225
    NAME: cosine
  MIN_LR: 5.0e-06
  OPTIMIZER:
    BETAS:
    - 0.9
    - 0.999
    EPS: 1.0e-08
    MOMENTUM: 0.9
    NAME: sgd
  START_EPOCH: 0
  USE_CHECKPOINT: false
  WARMUP_EPOCHS: 5
  WARMUP_LR: 5.0e-07
  WEIGHT_DECAY: 5.0e-05

Inference ``` cd CoModels/cv/classification/regnet_y_32gf bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/96552a7d-5080-4978-9fbb-52dc19dfac66)
推理结果 INFO * Acc@1 66.826Acc@5 71.245
kokuro-asahi commented 11 months ago

regnet_x_400mf

cd CoModels/cv/classification/regnet_x_400mf
bash train.sh
训练所用超参数 
AMP_OPT_LEVEL: ''
AUG:
  AUTO_AUGMENT: rand-m9-mstd0.5-inc1
  COLOR_JITTER: 0.4
  CUTMIX: 0.0
  CUTMIX_MINMAX: null
  MIXUP: 0.0
  MIXUP_MODE: batch
  MIXUP_PROB: 1.0
  MIXUP_SWITCH_PROB: 0.5
  RECOUNT: 1
  REMODE: pixel
  REPROB: 0.25
BASE:
- ''
DATA:
  BATCH_SIZE: 128
  CACHE_MODE: part
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224
  INTERPOLATION: bicubic
  NUM_CLASSES: 1000
  NUM_WORKERS: 4
  PIN_MEMORY: true
  SYNTHETIC_DATA: false
  ZIP_MODE: false
EVAL_MODE: true
LOCAL_RANK: 0
MODEL:
  ARCH: regnet_x_400mf
  CHECKPOINTS: null
  DROP_PATH_RATE: 0.1
  DROP_RATE: 0.0
  LABEL_SMOOTHING: 0.1
  NUM_CLASSES: 1000
  PRETRAINED: true
  RESUME: ''
OUTPUT: output/regnet_x_400mf/default
PRINT_FREQ: 50
SAVE_FREQ: 1
SEED: 42
TAG: default
TEST:
  CROP: true
  SEQUENTIAL: false
THROUGHPUT_MODE: false
TRAIN:
  ACCUMULATION_STEPS: 0
  AUTO_RESUME: false
  BASE_LR: 0.4
  CLIP_GRAD: 5.0
  EPOCHS: 100
  LR_SCHEDULER:
    DECAY_EPOCHS: 30
    DECAY_RATE: 0.1
    MILESTONES:
    - 150
    - 225
    NAME: cosine
  MIN_LR: 5.0e-06
  OPTIMIZER:
    BETAS:
    - 0.9
    - 0.999
    EPS: 1.0e-08
    MOMENTUM: 0.9
    NAME: sgd
  START_EPOCH: 0
  USE_CHECKPOINT: false
  WARMUP_EPOCHS: 5
  WARMUP_LR: 5.0e-07
  WEIGHT_DECAY: 5.0e-05

Inference ``` cd CoModels/cv/classification/regnet_x_400mf bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/59110141/a272c7e5-0168-47b7-bf7d-f57548b3ecae)
推理结果 INFO * Acc@1 76.530Acc@5 89.627
akeeei commented 11 months ago

deit_base_distilled_patch16_384

cd CoModels/cv/classification/deit_base_distilled_patch16_384
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 16
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 384

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/deit_base_distilled_patch16_384 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/c61cde5e-c5a5-4817-84e3-0d4ad03cc7f3) * 训练结果 ： ``` INFO * Acc@1 84.444 Acc@5 96.819 INFO Accuracy of the network on the 391 test images: 84.4% INFO Max accuracy: 84.44% INFO Training time 2:05:52 ```
推理结果 ``` INFO * Acc@1 74.549 Acc@5 91.115 INFO Accuracy of the network on the 1563 test images: 74.5% INFO throughput averaged with 30 times INFO batch_size 32 throughput 62.36388494704618 ```
akeeei commented 11 months ago

deit_base_distilled_patch16_224

cd CoModels/cv/classification/deit_base_distilled_patch16_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 32
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 5e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/deit_base_distilled_patch16_224 bash infer.sh ```
训练过程 * 训练日志 ： * ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/31f6a362-4a14-4bcf-8f02-7e49962c8b3e) * 训练结果 ： ``` INFO * Acc@1 82.906 Acc@5 96.165 INFO Accuracy of the network on the 391 test images: 82.9% INFO Max accuracy: 82.91% INFO Training time 1:54:26 ```
推理结果 ``` INFO * Acc@1 76.882 Acc@5 93.174 INFO Accuracy of the network on the 1563 test images: 76.9% INFO throughput averaged with 30 times INFO batch_size 32 throughput 135.36690275982912 ```
akeeei commented 11 months ago

deit_tiny_distilled_patch16_224

cd CoModels/cv/classification/deit_tiny_distilled_patch16_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 64
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 5e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/deit_tiny_distilled_patch16_224 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/a6a26ef4-9f9d-4119-aaa9-bbda5071865b) * 训练结果 ： ``` INFO * Acc@1 73.432 Acc@5 91.246 INFO Accuracy of the network on the 196 test images: 73.4% INFO Max accuracy: 73.43% INFO Training time 1:47:11 ```
推理结果 ``` INFO * Acc@1 62.288 Acc@5 83.706 INFO Accuracy of the network on the 1563 test images: 62.3% INFO throughput averaged with 30 times INFO batch_size 32 throughput 3396.701804586376 ```
akeeei commented 11 months ago

deit_small_distilled_patch16_224

cd CoModels/cv/classification/deit_small_distilled_patch16_224
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 64
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 5e-5
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/deit_small_distilled_patch16_224 bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/34331de0-05c7-4a6d-abe3-6407404e4851) * 训练结果 ： ``` INFO * Acc@1 80.637 Acc@5 95.155 INFO Accuracy of the network on the 196 test images: 80.6% INFO Max accuracy: 80.64% INFO Training time 0:48:11 ```
推理结果 ``` INFO * Acc@1 81.186 Acc@5 95.412 INFO Accuracy of the network on the 1563 test images: 81.2% INFO throughput averaged with 30 times INFO batch_size 32 throughput 1505.8929408904762 ```
akeeei commented 11 months ago

Fan hybrid base

cd CoModels/cv/classification/fan_hybrid_base
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 64
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/fan_hybrid_base bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/854f9be7-a362-4bef-9f4a-fe94b3a09e2e) * 训练结果 ： ``` INFO * Acc@1 82.906 Acc@5 96.165 INFO Accuracy of the network on the 391 test images: 82.9% INFO Max accuracy: 82.91% INFO Training time 1:54:26 ```
推理结果 ``` INFO * Acc@1 76.882 Acc@5 93.174 INFO Accuracy of the network on the 1563 test images: 76.9% INFO throughput averaged with 30 times INFO batch_size 32 throughput 135.36690275982912 ```
akeeei commented 11 months ago

Fan hybrid small

cd CoModels/cv/classification/fan_hybrid_small
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 64
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/fan_hybrid_small bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/7f0a891a-cac2-44b7-8b50-e7cce130af8e) * 训练结果 ： ``` INFO * Acc@1 83.216 Acc@5 96.420 INFO Accuracy of the network on the 196 test images: 83.2% INFO Max accuracy: 83.22% INFO Training time 2:29:48 ```
推理结果 ``` INFO * Acc@1 83.519 Acc@5 96.577 INFO Accuracy of the network on the 1563 test images: 83.5% INFO throughput averaged with 30 times INFO batch_size 32 throughput 169.14599193751516 ```
akeeei commented 11 months ago

Fan hybrid tiny

cd CoModels/cv/classification/fan_hybrid_tiny
bash train.sh
训练所用超参数 
DATA:
  BATCH_SIZE: 32
  DATASET: imagenet
  DATA_PATH: /data/dataset/ImageNet/extract
  IMG_SIZE: 224

MODEL:
  PRETRAINED: True
  LABEL_SMOOTHING: 0.11

TRAIN:
  START_EPOCH: 0
  EPOCHS: 90
  WARMUP_EPOCHS: 30
  WEIGHT_DECAY: 0.3
  BASE_LR: 1e-4
  WARMUP_LR: 5e-7

  LR_SCHEDULER:
    NAME: step

  OPTIMIZER:
    NAME: sgd
    MOMENTUM: 0.9

Inference ``` cd CoModels/cv/classification/fan_hybrid_tiny bash infer.sh ```
训练过程 * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/98879022/7e884073-e734-44eb-afa8-6e251195be7b) * 训练结果 ： ``` INFO * Acc@1 79.988 Acc@5 95.086 INFO Accuracy of the network on the 261 test images: 80.0% INFO Max accuracy: 79.99% INFO Training time 2:25:12 ```
推理结果 ``` INFO * Acc@1 80.098 Acc@5 95.068 INFO Accuracy of the network on the 1563 test images: 80.1% INFO throughput averaged with 30 times INFO batch_size 32 throughput 869.9123202721131 ```
iwkkk commented 11 months ago

mlp_mixer_l16_224

Training ``` cd CoModels/cv/classification/mlp_mixer_l16_224 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 32 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 8 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: mlp_mixer_l16_224 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/mlp_mixer_l16_224/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.001 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 1.25e-06 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/mlp_mixer_l16_224 bash infer.sh ```
训练过程（1 epoch） * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/4773d6df-d85c-4594-93e7-95959a6fb6af)
推理结果 INFO * Acc@1 68.654 Acc@5 86.131 INFO Accuracy of the network on the 6250 test images: 68.7%
iwkkk commented 11 months ago

convmixer_768_32_relu

Training ``` cd CoModels/cv/classification/convmixer_768_32_relu bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.5 CUTMIX_MINMAX: null MIXUP: 0.5 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 16 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bilinear NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: convmixer_768_32_relu CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/convmixer_768_32_relu/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.001 CLIP_GRAD: 1.0 EPOCHS: 150 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: cosine MIN_LR: 6.25e-07 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: adamw START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 2.0e-05 ```
Inference ``` cd CoModels/cv/classification/convmixer_768_32_relu bash infer.sh ```
训练过程（3 epochs） * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/aedcf136-e024-4679-9a8b-c2e3bd2b4544)
推理结果 INFO * Acc@1 80.081 Acc@5 94.995 INFO Accuracy of the network on the 1563 test images: 80.1%
iwkkk commented 11 months ago

convmixer_1536_20

Training ``` cd CoModels/cv/classification/convmixer_1536_20 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.5 CUTMIX_MINMAX: null MIXUP: 0.5 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 16 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 224 INTERPOLATION: bilinear NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: convmixer_1536_20 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/convmixer_1536_20/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.01 CLIP_GRAD: 1.0 EPOCHS: 150 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: cosine MIN_LR: 6.25e-07 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: adamw START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 2.0e-05 ```
Inference ``` cd CoModels/cv/classification/convmixer_1536_20 bash infer.sh ```
训练过程（2 epochs） * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/b8c4195c-e2af-4486-81ec-ed7a664a7eb8)
推理结果 INFO * Acc@1 81.039 Acc@5 95.610 INFO Accuracy of the network on the 6250 test images: 81.0%
iwkkk commented 11 months ago

cait_XS24_384

Training ``` cd CoModels/cv/classification/cait_XS24_384 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 8 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 384 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: cait_XS24_384 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/cait_XS24_384/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 1.0e-05 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 3.125e-07 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/cait_XS24_384 bash infer.sh ```
训练过程（1 epoch） * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/b081b0b1-4f46-445c-bf22-f41c53392dff)
推理结果 INFO * Acc@1 83.756 Acc@5 96.749 INFO Accuracy of the network on the 12500 test images: 83.8%
iwkkk commented 11 months ago

cait_M36_384

Training ``` cd CoModels/cv/classification/cait_M36_384 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 4 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 384 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: cait_M36_384 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/cait_M36_384/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 1.0e-05 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 1.5625e-07 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/cait_M36_384 bash infer.sh ```
训练过程（1 epoch） * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/c4005f41-7af1-488a-b954-e896d6702ad1)
推理结果 INFO * Acc@1 85.928 Acc@5 97.583 INFO Accuracy of the network on the 3125 test images: 85.9%
iwkkk commented 11 months ago

cait_S24_384

Training ``` cd CoModels/cv/classification/cait_S24_384 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 16 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 384 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 8 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: cait_S24_384 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/cait_S24_384/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 1.0e-05 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 6.25e-07 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/cait_S24_384 bash infer.sh ```
训练过程（1 epoch） ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/529e0f20-74a1-4e11-83d8-1122c29bfd3e) * 训练日志 ：
推理结果 INFO * Acc@1 84.956 Acc@5 97.219 INFO Accuracy of the network on the 3125 test images: 85.0%
iwkkk commented 11 months ago

cait_S36_384

Training ``` cd CoModels/cv/classification/cait_S36_384 bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 8 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 384 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: cait_S36_384 CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/cait_S36_384/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 1.0e-05 CLIP_GRAD: 5.0 EPOCHS: 90 LR_SCHEDULER: DECAY_EPOCHS: 30 DECAY_RATE: 0.1 MILESTONES: - 150 - 225 NAME: step MIN_LR: 3.125e-07 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/cait_S36_384 bash infer.sh ```
训练过程（1 epoch） * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/5ec8f2ed-59a4-4646-8df2-5b3ca6e0b164)
推理结果 INFO * Acc@1 85.196 Acc@5 97.386 INFO Accuracy of the network on the 3125 test images: 85.2%
iwkkk commented 11 months ago

genet_normal

Training ``` cd CoModels/cv/classification/genet_normal bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 16 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 192 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: genet_normal CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/genet_normal/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.01 CLIP_GRAD: 5.0 EPOCHS: 20 LR_SCHEDULER: DECAY_EPOCHS: 1 DECAY_RATE: 0.8 MILESTONES: - 150 - 225 NAME: step MIN_LR: 6.25e-07 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/genet_normal bash infer.sh ```
训练过程（1 epoch） * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/3069107e-c0ea-47e5-8d73-3eb613072ec1)
推理结果 INFO * Acc@1 79.610 Acc@5 94.500 INFO Accuracy of the network on the 12500 test images: 79.6%
iwkkk commented 11 months ago

genet_small

Training ``` cd CoModels/cv/classification/genet_small bash train.sh ``` 训练所用超参数 ``` AMP_OPT_LEVEL: '' AUG: AUTO_AUGMENT: rand-m9-mstd0.5-inc1 COLOR_JITTER: 0.4 CUTMIX: 0.0 CUTMIX_MINMAX: null MIXUP: 0.0 MIXUP_MODE: batch MIXUP_PROB: 1.0 MIXUP_SWITCH_PROB: 0.5 RECOUNT: 1 REMODE: pixel REPROB: 0.25 BASE: - '' DATA: BATCH_SIZE: 16 CACHE_MODE: part DATASET: imagenet DATA_PATH: /data/dataset/ImageNet/extract IMG_SIZE: 192 INTERPOLATION: bicubic NUM_CLASSES: 1000 NUM_WORKERS: 4 PIN_MEMORY: true SYNTHETIC_DATA: false ZIP_MODE: false EVAL_MODE: true LOCAL_RANK: 0 MODEL: ARCH: genet_small CHECKPOINTS: null DROP_PATH_RATE: 0.1 DROP_RATE: 0.0 LABEL_SMOOTHING: 0.1 NUM_CLASSES: 1000 PRETRAINED: true RESUME: '' OUTPUT: output/genet_small/default PRINT_FREQ: 50 SAVE_FREQ: 1 SEED: 42 TAG: default TEST: CROP: true SEQUENTIAL: false THROUGHPUT_MODE: false TRAIN: ACCUMULATION_STEPS: 0 AUTO_RESUME: false BASE_LR: 0.01 CLIP_GRAD: 5.0 EPOCHS: 20 LR_SCHEDULER: DECAY_EPOCHS: 1 DECAY_RATE: 0.8 MILESTONES: - 150 - 225 NAME: step MIN_LR: 6.25e-07 OPTIMIZER: BETAS: - 0.9 - 0.999 EPS: 1.0e-08 MOMENTUM: 0.9 NAME: sgd START_EPOCH: 0 USE_CHECKPOINT: false WARMUP_EPOCHS: 0 WARMUP_LR: 5.0e-07 WEIGHT_DECAY: 0.0001 ```
Inference ``` cd CoModels/cv/classification/genet_small bash infer.sh ```
训练过程（1 epoch） * 训练日志 ： ![image](https://github.com/Oneflow-Inc/CoModels/assets/77448166/52fda8c9-5bda-40aa-bd7e-0ecb4777c664)
推理结果 INFO * Acc@1 75.321 Acc@5 92.252 INFO Accuracy of the network on the 12500 test images: 75.3%