deepvac_clothes_accept600 SOTA

[gemfield]

IoU 金榜

• test5: 58.6% (62.6%)
• test6: 58.4% (62.7%)
• test4: 57.8% (62.4%)
• test3: 57.4% (61.5%)
• test2: 57.2% (61.5%)
• test1: 53.6% (57.1%)

[buptlihang]

test1

model

model_esp_epoch_248_miou_0.7251.pth

IOU

• images: 0.536
• images_no_dress: 0.571
• val data: 0.7251

特别超参

• 训练集：std2.1
• 网络：original ESPNetV2
• config.input_w = 384
• config.input_h = 384
• config.core.optimizer = torch.optim.SGD(config.core.net.parameters(), 5e-3, momentum=0.9) config.core.scheduler = optim.lr_scheduler.MultiStepLR(config.core.optimizer, milestones=[20,40,55,70,80,90,100,110,120,130,140,150,160], gamma=0.27030)

基本超参

• config.core.cls_num = 4
• config.core.batch_size = 16
• config.core.mean = config.data['mean']
• config.core.std = config.data['std']
• config.core.model_path = "/opt/public/pretrain/ESPNetv2/imagenet/espnetv2_s_2.0.pth"
• config.core.criterion = torch.nn.CrossEntropyLoss(weight)
• BorderTargetAugFactory('SpeckleAug@0.1 => GaussianAug@0.1 => HorlineAug@0.1 => VerlineAug@0.1 => LRmotionAug@0.1 => UDmotionAug@0.1 => NoisyAug@0.1 => DarkAug@0.1 => ColorJitterAug@0.25 => BrightnessJitterAug@0.25 => ContrastJitterAug@0.25 => ImageWithMasksRandomRotateAug@0.6 => ImageWithMasksCenterCropAug => ImageWithMasksScaleAug => ImageWithMasksHFlipAug@0.5 => ImageWithMasksNormalizeAug => BorderTargetAug => ImageWithTwoMasksToTensorAug', deepvac_config)

[buptlihang]

test2

model

gemfield_epoch_186.pth

IOU

• images: 0.572
• images_no_dress: 0.615
• val data: 0.73?

特别超参

• 训练集：std2.1
• 网络：original ESPNetV2
• config.input_w = 384
• config.input_h = 384
• config.core.optimizer = torch.optim.Adam(config.core.net.parameters(), 5e-4, (0.9, 0.999)) lambda_lr = lambda epoch: round ((1 - epoch/config.core.epoch_num) ** 0.9, 8) config.core.scheduler = optim.lr_scheduler.LambdaLR(config.core.optimizer, lr_lambda=lambda_lr)

基本超参

• config.core.cls_num = 4
• config.core.batch_size = 16
• config.core.mean = config.data['mean']
• config.core.std = config.data['std']
• config.core.model_path = "/opt/public/pretrain/ESPNetv2/imagenet/espnetv2_s_2.0.pth"
• config.core.criterion = torch.nn.CrossEntropyLoss(weight)
• BorderTargetAugFactory('SpeckleAug@0.1 => GaussianAug@0.1 => HorlineAug@0.1 => VerlineAug@0.1 => LRmotionAug@0.1 => UDmotionAug@0.1 => NoisyAug@0.1 => DarkAug@0.1 => ColorJitterAug@0.25 => BrightnessJitterAug@0.25 => ContrastJitterAug@0.25 => ImageWithMasksRandomRotateAug@0.6 => ImageWithMasksCenterCropAug => ImageWithMasksScaleAug => ImageWithMasksHFlipAug@0.5 => ImageWithMasksNormalizeAug => BorderTargetAug => ImageWithTwoMasksToTensorAug', deepvac_config)

[buptlihang]

test3

model

model_fusion_0.7428.pth

IOU

• images：0.574
• images_no_dress: 0.615
• val data: 0.7428

特别超参

• 训练集：std2.1
• 网络：semantic+detail+fusion
• config.input_w = 384
• config.input_h = 384
• config.core.optimizer = torch.optim.SGD(config.core.net.parameters(), 5e-3, momentum=0.9)
• config.core.scheduler = optim.lr_scheduler.MultiStepLR(config.core.optimizer, milestones=[20,40,55,70,80,90,100,110,120,130,140,150,160], gamma=0.27030)

基本超参

• config.core.cls_num = 4
• config.core.batch_size = 16
• config.core.mean = config.data['mean']
• config.core.std = config.data['std']
• config.core.model_path = "/opt/public/pretrain/ESPNetv2/imagenet/espnetv2_s_2.0.pth"
• config.core.criterion = torch.nn.CrossEntropyLoss(weight)
• BorderTargetAugFactory('SpeckleAug@0.1 => GaussianAug@0.1 => HorlineAug@0.1 => VerlineAug@0.1 => LRmotionAug@0.1 => UDmotionAug@0.1 \ => NoisyAug@0.1 => DarkAug@0.1 => ColorJitterAug@0.25 => BrightnessJitterAug@0.25 => ContrastJitterAug@0.25 => \ ImageWithMasksRandomRotateAug@0.6 => ImageWithMasksCenterCropAug => ImageWithMasksScaleAug => ImageWithMasksHFlipAug@0.5 => \ ImageWithMasksNormalizeAug => BorderTargetAug => ImageWithTwoMasksToTensorAug', deepvac_config)

[buptlihang]

test4

model

model_fusion_0.7473.pth

IOU

• images: 0.578
• images_no_dress: 0.624
• val data: 0.7473

特别超参

• 训练集：std2.1
• 网络：semantic+detail+fusion
• config.input_w = 384
• config.input_h = 384
• config.core.optimizer = torch.optim.Adam(config.core.net.parameters(), 5e-4, (0.9, 0.999)) lambda_lr = lambda epoch: round ((1 - epoch/config.core.epoch_num) ** 0.9, 8) config.core.scheduler = optim.lr_scheduler.LambdaLR(config.core.optimizer, lr_lambda=lambda_lr)

基本超参

• config.core.cls_num = 4
• config.core.batch_size = 16
• config.core.mean = config.data['mean']
• config.core.std = config.data['std']
• config.core.model_path = "/opt/public/pretrain/ESPNetv2/imagenet/espnetv2_s_2.0.pth"
• config.core.criterion = torch.nn.CrossEntropyLoss(weight)
• BorderTargetAugFactory('SpeckleAug@0.1 => GaussianAug@0.1 => HorlineAug@0.1 => VerlineAug@0.1 => LRmotionAug@0.1 => UDmotionAug@0.1 => NoisyAug@0.1 => DarkAug@0.1 => ColorJitterAug@0.25 => BrightnessJitterAug@0.25 => ContrastJitterAug@0.25 => ImageWithMasksRandomRotateAug@0.6 => ImageWithMasksCenterCropAug => ImageWithMasksScaleAug => ImageWithMasksHFlipAug@0.5 => ImageWithMasksNormalizeAug => BorderTargetAug => ImageWithTwoMasksToTensorAug', deepvac_config)

[buptlihang]

test5

model

model_synthesis_epoch_155_0.7394.pth

IOU

• images: 0.586
• images_no_dress: 0.626
• val data: 0.7394

特别超参

• 训练集：std2.1+synthesis(2800+5700)
• 网络：original ESPNetV2
• config.input_w = 384
• config.input_h = 384
• config.core.optimizer = torch.optim.SGD(config.core.net.parameters(), 5e-3, momentum=0.9) config.core.scheduler = optim.lr_scheduler.MultiStepLR(config.core.optimizer, milestones=[20,40,55,70,80,90,100,110,120,130,140,150,160], gamma=0.27030)

基本超参

• config.core.cls_num = 4
• config.core.batch_size = 16
• config.core.mean = config.data['mean']
• config.core.std = config.data['std']
• config.core.model_path = "/opt/public/pretrain/ESPNetv2/imagenet/espnetv2_s_2.0.pth"
• config.core.criterion = torch.nn.CrossEntropyLoss(weight)
• BorderTargetAugFactory('SpeckleAug@0.1 => GaussianAug@0.1 => HorlineAug@0.1 => VerlineAug@0.1 => LRmotionAug@0.1 => UDmotionAug@0.1 => NoisyAug@0.1 => DarkAug@0.1 => ColorJitterAug@0.25 => BrightnessJitterAug@0.25 => ContrastJitterAug@0.25 => ImageWithMasksRandomRotateAug@0.6 => ImageWithMasksCenterCropAug => ImageWithMasksScaleAug => ImageWithMasksHFlipAug@0.5 => ImageWithMasksNormalizeAug => BorderTargetAug => ImageWithTwoMasksToTensorAug', deepvac_config)

[buptlihang]

test6

model

model_synthesis_bg_0.7405.pth

IOU

• images: 0.584
• images_no_dress: 0.627
• val data: 0.7405

特别超参

• 训练集：std2.1+Synthesis+bg(2800+5700+2000)
• 网络：original ESPNetV2
• config.input_w = 384
• config.input_h = 384
• config.core.optimizer = torch.optim.SGD(config.core.net.parameters(), 5e-3, momentum=0.9) config.core.scheduler = optim.lr_scheduler.MultiStepLR(config.core.optimizer, milestones=[20,40,55,70,80,90,100,110,120,130,140,150,160], gamma=0.27030)

基本超参

• config.core.cls_num = 4
• config.core.batch_size = 16
• config.core.mean = config.data['mean']
• config.core.std = config.data['std']
• config.core.model_path = "/opt/public/pretrain/ESPNetv2/imagenet/espnetv2_s_2.0.pth"
• config.core.criterion = torch.nn.CrossEntropyLoss(weight)
• BorderTargetAugFactory('SpeckleAug@0.1 => GaussianAug@0.1 => HorlineAug@0.1 => VerlineAug@0.1 => LRmotionAug@0.1 => UDmotionAug@0.1 => NoisyAug@0.1 => DarkAug@0.1 => ColorJitterAug@0.25 => BrightnessJitterAug@0.25 => ContrastJitterAug@0.25 => ImageWithMasksRandomRotateAug@0.6 => ImageWithMasksCenterCropAug => ImageWithMasksScaleAug => ImageWithMasksHFlipAug@0.5 => ImageWithMasksNormalizeAug => BorderTargetAug => ImageWithTwoMasksToTensorAug', deepvac_config)