search

HRNet / HRNet-Semantic-Segmentation

The OCR approach is rephrased as Segmentation Transformer: https://arxiv.org/abs/1909.11065. This is an official implementation of semantic segmentation for HRNet. https://arxiv.org/abs/1908.07919
Other
3.09k stars 683 forks source link

lower result than using default config #203

Open purse1996 opened 3 years ago

purse1996 commented 3 years ago

When I run the code using default config file, the result is 80.99, which is lower than 81.6. Can you offer some suggestions? Seeding with 304 => creating output/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484 => creating log/cityscapes/seg_hrnet_ocr/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484_2020-11-28-17-18 Namespace(cfg='experiments/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484.yaml', local_rank=3, opts=[], seed=304) AUTO_RESUME: False CUDNN: BENCHMARK: True DETERMINISTIC: False ENABLED: True DATASET: DATASET: cityscapes EXTRA_TRAIN_SET: NUM_CLASSES: 19 ROOT: data/ TEST_SET: list/cityscapes/val.lst TRAIN_SET: list/cityscapes/train.lst DEBUG: DEBUG: False SAVE_BATCH_IMAGES_GT: False SAVE_BATCH_IMAGES_PRED: False SAVE_HEATMAPS_GT: False SAVE_HEATMAPS_PRED: False GPUS: (0, 1, 2, 3) LOG_DIR: log LOSS: BALANCE_WEIGHTS: [0.4, 1] CLASS_BALANCE: False OHEMKEEP: 131072 OHEMTHRES: 0.9 USE_OHEM: False MODEL: ALIGN_CORNERS: True EXTRA: FINAL_CONV_KERNEL: 1 STAGE1: BLOCK: BOTTLENECK FUSE_METHOD: SUM NUM_BLOCKS: [4] NUM_CHANNELS: [64] NUM_MODULES: 1 NUM_RANCHES: 1 STAGE2: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4] NUM_BRANCHES: 2 NUM_CHANNELS: [48, 96] NUM_MODULES: 1 STAGE3: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4, 4] NUM_BRANCHES: 3 NUM_CHANNELS: [48, 96, 192] NUM_MODULES: 4 STAGE4: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4, 4, 4] NUM_BRANCHES: 4 NUM_CHANNELS: [48, 96, 192, 384] NUM_MODULES: 3 NAME: seg_hrnet_ocr NUM_OUTPUTS: 2 OCR: DROPOUT: 0.05 KEY_CHANNELS: 256 MID_CHANNELS: 512 SCALE: 1 PRETRAINED: pretrained_models/hrnetv2_w48_imagenet_pretrained.pth OUTPUT_DIR: output PIN_MEMORY: True PRINT_FREQ: 10 RANK: 0 TEST: BASE_SIZE: 2048 BATCH_SIZE_PER_GPU: 4 FLIP_TEST: False IMAGE_SIZE: [2048, 1024] MODEL_FILE: MULTI_SCALE: False NUM_SAMPLES: 0 OUTPUT_INDEX: -1 SCALE_LIST: [1] TRAIN: BASE_SIZE: 2048 BATCH_SIZE_PER_GPU: 3 BEGIN_EPOCH: 0 DOWNSAMPLERATE: 1 END_EPOCH: 484 EXTRA_EPOCH: 0 EXTRA_LR: 0.001 FLIP: True FREEZE_EPOCHS: -1 FREEZE_LAYERS: IGNORE_LABEL: 255 IMAGE_SIZE: [1024, 512] LR: 0.01 LR_FACTOR: 0.1 LR_STEP: [90, 110] MOMENTUM: 0.9 MULTI_SCALE: True NESTEROV: False NONBACKBONE_KEYWORDS: [] NONBACKBONE_MULT: 10 NUM_SAMPLES: 0 OPTIMIZER: sgd RANDOM_BRIGHTNESS: False RANDOM_BRIGHTNESS_SHIFT_VALUE: 10 RESUME: True SCALE_FACTOR: 16 SHUFFLE: True WD: 0.0005 WORKERS: 4 Seeding with 304 => creating output/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484 => creating log/cityscapes/seg_hrnet_ocr/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484_2020-11-28-17-18 Namespace(cfg='experiments/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484.yaml', local_rank=1, opts=[], seed=304) AUTO_RESUME: False CUDNN: BENCHMARK: True DETERMINISTIC: False ENABLED: True DATASET: DATASET: cityscapes EXTRA_TRAIN_SET: NUM_CLASSES: 19 ROOT: data/ TEST_SET: list/cityscapes/val.lst TRAIN_SET: list/cityscapes/train.lst DEBUG: DEBUG: False SAVE_BATCH_IMAGES_GT: False SAVE_BATCH_IMAGES_PRED: False SAVE_HEATMAPS_GT: False SAVE_HEATMAPS_PRED: False GPUS: (0, 1, 2, 3) LOG_DIR: log LOSS: BALANCE_WEIGHTS: [0.4, 1] CLASS_BALANCE: False OHEMKEEP: 131072 OHEMTHRES: 0.9 USE_OHEM: False MODEL: ALIGN_CORNERS: True EXTRA: FINAL_CONV_KERNEL: 1 STAGE1: BLOCK: BOTTLENECK FUSE_METHOD: SUM NUM_BLOCKS: [4] NUM_CHANNELS: [64] NUM_MODULES: 1 NUM_RANCHES: 1 STAGE2: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4] NUM_BRANCHES: 2 NUM_CHANNELS: [48, 96] NUM_MODULES: 1 STAGE3: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4, 4] NUM_BRANCHES: 3 NUM_CHANNELS: [48, 96, 192] NUM_MODULES: 4 STAGE4: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4, 4, 4] NUM_BRANCHES: 4 NUM_CHANNELS: [48, 96, 192, 384] NUM_MODULES: 3 NAME: seg_hrnet_ocr NUM_OUTPUTS: 2 OCR: DROPOUT: 0.05 KEY_CHANNELS: 256 MID_CHANNELS: 512 SCALE: 1 PRETRAINED: pretrained_models/hrnetv2_w48_imagenet_pretrained.pth OUTPUT_DIR: output PIN_MEMORY: True PRINT_FREQ: 10 RANK: 0 TEST: BASE_SIZE: 2048 BATCH_SIZE_PER_GPU: 4 FLIP_TEST: False IMAGE_SIZE: [2048, 1024] MODEL_FILE: MULTI_SCALE: False NUM_SAMPLES: 0 OUTPUT_INDEX: -1 SCALE_LIST: [1] TRAIN: BASE_SIZE: 2048 BATCH_SIZE_PER_GPU: 3 BEGIN_EPOCH: 0 DOWNSAMPLERATE: 1 END_EPOCH: 484 EXTRA_EPOCH: 0 EXTRA_LR: 0.001 FLIP: True FREEZE_EPOCHS: -1 FREEZE_LAYERS: IGNORE_LABEL: 255 IMAGE_SIZE: [1024, 512] LR: 0.01 LR_FACTOR: 0.1 LR_STEP: [90, 110] MOMENTUM: 0.9 MULTI_SCALE: True NESTEROV: False NONBACKBONE_KEYWORDS: [] NONBACKBONE_MULT: 10 NUM_SAMPLES: 0 OPTIMIZER: sgd RANDOM_BRIGHTNESS: False RANDOM_BRIGHTNESS_SHIFT_VALUE: 10 RESUME: True SCALE_FACTOR: 16 SHUFFLE: True WD: 0.0005 WORKERS: 4 Seeding with 304 => creating output/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484 => creating log/cityscapes/seg_hrnet_ocr/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484_2020-11-28-17-18 Namespace(cfg='experiments/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484.yaml', local_rank=2, opts=[], seed=304) AUTO_RESUME: False CUDNN: BENCHMARK: True DETERMINISTIC: False ENABLED: True DATASET: DATASET: cityscapes EXTRA_TRAIN_SET: NUM_CLASSES: 19 ROOT: data/ TEST_SET: list/cityscapes/val.lst TRAIN_SET: list/cityscapes/train.lst DEBUG: DEBUG: False SAVE_BATCH_IMAGES_GT: False SAVE_BATCH_IMAGES_PRED: False SAVE_HEATMAPS_GT: False SAVE_HEATMAPS_PRED: False GPUS: (0, 1, 2, 3) LOG_DIR: log LOSS: BALANCE_WEIGHTS: [0.4, 1] CLASS_BALANCE: False OHEMKEEP: 131072 OHEMTHRES: 0.9 USE_OHEM: False MODEL: ALIGN_CORNERS: True EXTRA: FINAL_CONV_KERNEL: 1 STAGE1: BLOCK: BOTTLENECK FUSE_METHOD: SUM NUM_BLOCKS: [4] NUM_CHANNELS: [64] NUM_MODULES: 1 NUM_RANCHES: 1 STAGE2: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4] NUM_BRANCHES: 2 NUM_CHANNELS: [48, 96] NUM_MODULES: 1 STAGE3: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4, 4] NUM_BRANCHES: 3 NUM_CHANNELS: [48, 96, 192] NUM_MODULES: 4 STAGE4: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4, 4, 4] NUM_BRANCHES: 4 NUM_CHANNELS: [48, 96, 192, 384] NUM_MODULES: 3 NAME: seg_hrnet_ocr NUM_OUTPUTS: 2 OCR: DROPOUT: 0.05 KEY_CHANNELS: 256 MID_CHANNELS: 512 SCALE: 1 PRETRAINED: pretrained_models/hrnetv2_w48_imagenet_pretrained.pth OUTPUT_DIR: output PIN_MEMORY: True PRINT_FREQ: 10 RANK: 0 TEST: BASE_SIZE: 2048 BATCH_SIZE_PER_GPU: 4 FLIP_TEST: False IMAGE_SIZE: [2048, 1024] MODEL_FILE: MULTI_SCALE: False NUM_SAMPLES: 0 OUTPUT_INDEX: -1 SCALE_LIST: [1] TRAIN: BASE_SIZE: 2048 BATCH_SIZE_PER_GPU: 3 BEGIN_EPOCH: 0 DOWNSAMPLERATE: 1 END_EPOCH: 484 EXTRA_EPOCH: 0 EXTRA_LR: 0.001 FLIP: True FREEZE_EPOCHS: -1 FREEZE_LAYERS: IGNORE_LABEL: 255 IMAGE_SIZE: [1024, 512] LR: 0.01 LR_FACTOR: 0.1 LR_STEP: [90, 110] MOMENTUM: 0.9 MULTI_SCALE: True NESTEROV: False NONBACKBONE_KEYWORDS: [] NONBACKBONE_MULT: 10 NUM_SAMPLES: 0 OPTIMIZER: sgd RANDOM_BRIGHTNESS: False RANDOM_BRIGHTNESS_SHIFT_VALUE: 10 RESUME: True SCALE_FACTOR: 16 SHUFFLE: True WD: 0.0005 WORKERS: 4 Seeding with 304 => creating output/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484 => creating log/cityscapes/seg_hrnet_ocr/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484_2020-11-28-17-18 Namespace(cfg='experiments/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484.yaml', local_rank=0, opts=[], seed=304) AUTO_RESUME: False CUDNN: BENCHMARK: True DETERMINISTIC: False ENABLED: True DATASET: DATASET: cityscapes EXTRA_TRAIN_SET: NUM_CLASSES: 19 ROOT: data/ TEST_SET: list/cityscapes/val.lst TRAIN_SET: list/cityscapes/train.lst DEBUG: DEBUG: False SAVE_BATCH_IMAGES_GT: False SAVE_BATCH_IMAGES_PRED: False SAVE_HEATMAPS_GT: False SAVE_HEATMAPS_PRED: False GPUS: (0, 1, 2, 3) LOG_DIR: log LOSS: BALANCE_WEIGHTS: [0.4, 1] CLASS_BALANCE: False OHEMKEEP: 131072 OHEMTHRES: 0.9 USE_OHEM: False MODEL: ALIGN_CORNERS: True EXTRA: FINAL_CONV_KERNEL: 1 STAGE1: BLOCK: BOTTLENECK FUSE_METHOD: SUM NUM_BLOCKS: [4] NUM_CHANNELS: [64] NUM_MODULES: 1 NUM_RANCHES: 1 STAGE2: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4] NUM_BRANCHES: 2 NUM_CHANNELS: [48, 96] NUM_MODULES: 1 STAGE3: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4, 4] NUM_BRANCHES: 3 NUM_CHANNELS: [48, 96, 192] NUM_MODULES: 4 STAGE4: BLOCK: BASIC FUSE_METHOD: SUM NUM_BLOCKS: [4, 4, 4, 4] NUM_BRANCHES: 4 NUM_CHANNELS: [48, 96, 192, 384] NUM_MODULES: 3 NAME: seg_hrnet_ocr NUM_OUTPUTS: 2 OCR: DROPOUT: 0.05 KEY_CHANNELS: 256 MID_CHANNELS: 512 SCALE: 1 PRETRAINED: pretrained_models/hrnetv2_w48_imagenet_pretrained.pth OUTPUT_DIR: output PIN_MEMORY: True PRINT_FREQ: 10 RANK: 0 TEST: BASE_SIZE: 2048 BATCH_SIZE_PER_GPU: 4 FLIP_TEST: False IMAGE_SIZE: [2048, 1024] MODEL_FILE: MULTI_SCALE: False NUM_SAMPLES: 0 OUTPUT_INDEX: -1 SCALE_LIST: [1] TRAIN: BASE_SIZE: 2048 BATCH_SIZE_PER_GPU: 3 BEGIN_EPOCH: 0 DOWNSAMPLERATE: 1 END_EPOCH: 484 EXTRA_EPOCH: 0 EXTRA_LR: 0.001 FLIP: True FREEZE_EPOCHS: -1 FREEZE_LAYERS: IGNORE_LABEL: 255 IMAGE_SIZE: [1024, 512] LR: 0.01 LR_FACTOR: 0.1 LR_STEP: [90, 110] MOMENTUM: 0.9 MULTI_SCALE: True NESTEROV: False NONBACKBONE_KEYWORDS: [] NONBACKBONE_MULT: 10 NUM_SAMPLES: 0 OPTIMIZER: sgd RANDOM_BRIGHTNESS: False RANDOM_BRIGHTNESS_SHIFT_VALUE: 10 RESUME: True SCALE_FACTOR: 16 SHUFFLE: True WD: 0.0005 WORKERS: 4 => init weights from normal distribution => init weights from normal distribution => init weights from normal distribution => init weights from normal distribution => loading pretrained model pretrained_models/hrnetv2_w48_imagenet_pretrained.pth => loading pretrained model pretrained_models/hrnetv2_w48_imagenet_pretrained.pth {'ocr_distri_head.object_context_block.f_down.1.0.running_mean', 'ocr_distri_head.object_context_block.f_object.1.0.num_batches_tracked', 'cls_head.bias', 'ocr_distri_head.object_context_block.f_object.0.weight', 'ocr_distri_head.object_context_block.f_object.3.0.weight', 'ocr_distri_head.object_context_block.f_pixel.2.weight', 'ocr_distri_head.object_context_block.f_pixel.0.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.num_batches_tracked', 'conv3x3_ocr.1.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.running_var', 'ocr_distri_head.object_context_block.f_object.3.0.bias', 'ocr_distri_head.object_context_block.f_pixel.1.0.bias', 'ocr_distri_head.object_context_block.f_object.3.0.running_var', 'ocr_distri_head.object_context_block.f_object.3.0.num_batches_tracked', 'ocr_distri_head.conv_bn_dropout.0.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.bias', 'aux_head.1.weight', 'aux_head.0.weight', 'ocr_distri_head.object_context_block.f_down.1.0.running_var', 'ocr_distri_head.object_context_block.f_up.1.0.weight', 'ocr_distri_head.object_context_block.f_pixel.1.0.running_var', 'aux_head.1.running_var', 'ocr_distri_head.object_context_block.f_pixel.1.0.running_mean', 'ocr_distri_head.object_context_block.f_down.1.0.weight', 'ocr_distri_head.object_context_block.f_pixel.1.0.num_batches_tracked', 'ocr_distri_head.conv_bn_dropout.1.0.bias', 'ocr_distri_head.object_context_block.f_pixel.3.0.running_mean', 'ocr_distri_head.object_context_block.f_pixel.1.0.weight', 'ocr_distri_head.object_context_block.f_up.0.weight', 'conv3x3_ocr.0.weight', 'aux_head.1.running_mean', 'aux_head.1.num_batches_tracked', 'conv3x3_ocr.1.running_var', 'ocr_distri_head.object_context_block.f_object.3.0.running_mean', 'conv3x3_ocr.1.num_batches_tracked', 'ocr_distri_head.object_context_block.f_pixel.3.0.weight', 'ocr_distri_head.object_context_block.f_object.2.weight', 'aux_head.1.bias', 'ocr_distri_head.object_context_block.f_up.1.0.running_var', 'ocr_distri_head.conv_bn_dropout.1.0.running_var', 'ocr_distri_head.conv_bn_dropout.1.0.running_mean', 'ocr_distri_head.object_context_block.f_object.1.0.weight', 'conv3x3_ocr.1.running_mean', 'ocr_distri_head.object_context_block.f_up.1.0.bias', 'cls_head.weight', 'ocr_distri_head.object_context_block.f_object.1.0.running_mean', 'ocr_distri_head.conv_bn_dropout.1.0.num_batches_tracked', 'aux_head.3.weight', 'conv3x3_ocr.0.bias', 'ocr_distri_head.object_context_block.f_up.1.0.running_mean', 'aux_head.0.bias', 'ocr_distri_head.object_context_block.f_down.1.0.bias', 'ocr_distri_head.object_context_block.f_down.1.0.num_batches_tracked', 'conv3x3_ocr.1.bias', 'aux_head.3.bias', 'ocr_distri_head.object_context_block.f_up.1.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_object.1.0.running_var', 'ocr_distri_head.object_context_block.f_object.1.0.bias', 'ocr_distri_head.conv_bn_dropout.1.0.weight', 'ocr_distri_head.object_context_block.f_down.0.weight'} {'downsamp_modules.2.1.num_batches_tracked', 'incre_modules.3.0.conv3.weight', 'incre_modules.3.0.bn1.running_mean', 'incre_modules.2.0.downsample.1.bias', 'downsamp_modules.1.1.weight', 'downsamp_modules.2.1.running_var', 'incre_modules.1.0.bn1.bias', 'incre_modules.0.0.bn2.running_var', 'incre_modules.0.0.bn2.num_batches_tracked', 'incre_modules.2.0.conv3.weight', 'incre_modules.2.0.downsample.1.running_mean', 'incre_modules.2.0.bn2.num_batches_tracked', 'incre_modules.0.0.downsample.0.weight', 'incre_modules.1.0.bn3.running_mean', 'incre_modules.3.0.bn2.weight', 'incre_modules.1.0.bn1.running_var', 'incre_modules.3.0.bn3.num_batches_tracked', 'incre_modules.2.0.bn2.running_mean', 'incre_modules.3.0.bn2.running_var', 'incre_modules.2.0.bn3.num_batches_tracked', 'incre_modules.2.0.bn3.weight', 'downsamp_modules.2.1.weight', 'incre_modules.3.0.bn2.bias', 'downsamp_modules.1.1.bias', 'downsamp_modules.0.1.running_mean', 'incre_modules.1.0.bn2.num_batches_tracked', 'incre_modules.1.0.bn3.running_var', 'incre_modules.2.0.bn3.running_var', 'downsamp_modules.1.1.num_batches_tracked', 'downsamp_modules.0.1.bias', 'incre_modules.0.0.bn3.weight', 'incre_modules.2.0.bn2.weight', 'incre_modules.3.0.downsample.1.running_var', 'final_layer.1.num_batches_tracked', 'incre_modules.3.0.bn2.running_mean', 'downsamp_modules.0.1.running_var', 'downsamp_modules.2.1.running_mean', 'incre_modules.2.0.conv2.weight', 'incre_modules.0.0.downsample.1.num_batches_tracked', 'downsamp_modules.1.1.running_var', 'incre_modules.0.0.bn3.num_batches_tracked', 'downsamp_modules.1.1.running_mean', 'incre_modules.2.0.conv1.weight', 'incre_modules.3.0.conv2.weight', 'incre_modules.1.0.downsample.1.running_var', 'incre_modules.1.0.downsample.0.weight', 'incre_modules.0.0.bn3.bias', 'incre_modules.2.0.bn1.running_mean', 'incre_modules.2.0.downsample.1.num_batches_tracked', 'incre_modules.3.0.conv1.weight', 'incre_modules.3.0.downsample.1.weight', 'incre_modules.3.0.bn1.bias', 'incre_modules.1.0.conv2.weight', 'incre_modules.3.0.downsample.1.running_mean', 'incre_modules.1.0.bn2.weight', 'downsamp_modules.2.0.weight', 'incre_modules.2.0.bn3.bias', 'incre_modules.1.0.bn1.running_mean', 'incre_modules.0.0.conv2.weight', 'downsamp_modules.2.0.bias', 'final_layer.1.weight', 'incre_modules.1.0.bn2.bias', 'downsamp_modules.1.0.weight', 'downsamp_modules.2.1.bias', 'incre_modules.0.0.bn2.bias', 'incre_modules.0.0.bn1.running_var', 'downsamp_modules.0.0.weight', 'final_layer.1.running_mean', 'incre_modules.1.0.conv3.weight', 'incre_modules.2.0.bn2.bias', 'classifier.weight', 'incre_modules.2.0.bn3.running_mean', 'incre_modules.1.0.downsample.1.running_mean', 'incre_modules.0.0.downsample.1.running_mean', 'incre_modules.3.0.bn1.weight', 'incre_modules.3.0.downsample.0.weight', 'incre_modules.0.0.bn3.running_mean', 'incre_modules.0.0.bn1.weight', 'incre_modules.2.0.downsample.1.running_var', 'incre_modules.0.0.downsample.1.weight', 'final_layer.1.running_var', 'downsamp_modules.0.0.bias', 'final_layer.1.bias', 'incre_modules.2.0.bn1.bias', 'incre_modules.1.0.bn1.weight', 'incre_modules.2.0.bn1.weight', 'incre_modules.3.0.downsample.1.num_batches_tracked', 'incre_modules.0.0.conv1.weight', 'classifier.bias', 'final_layer.0.weight', 'incre_modules.0.0.bn3.running_var', 'incre_modules.1.0.downsample.1.weight', 'incre_modules.1.0.bn3.weight', 'incre_modules.0.0.downsample.1.running_var', 'incre_modules.1.0.conv1.weight', 'incre_modules.0.0.bn2.weight', 'incre_modules.2.0.bn1.num_batches_tracked', 'incre_modules.0.0.bn1.num_batches_tracked', 'incre_modules.3.0.bn3.running_mean', 'incre_modules.0.0.conv3.weight', 'incre_modules.2.0.bn1.running_var', 'incre_modules.3.0.bn3.weight', 'incre_modules.0.0.bn2.running_mean', 'incre_modules.1.0.bn1.num_batches_tracked', 'incre_modules.0.0.bn1.bias', 'incre_modules.3.0.bn2.num_batches_tracked', 'final_layer.0.bias', 'incre_modules.1.0.bn2.running_var', 'incre_modules.1.0.bn3.num_batches_tracked', 'incre_modules.1.0.bn3.bias', 'incre_modules.1.0.downsample.1.bias', 'incre_modules.3.0.bn3.bias', 'incre_modules.2.0.downsample.1.weight', 'incre_modules.0.0.bn1.running_mean', 'downsamp_modules.0.1.num_batches_tracked', 'downsamp_modules.0.1.weight', 'incre_modules.3.0.bn1.num_batches_tracked', 'incre_modules.0.0.downsample.1.bias', 'incre_modules.3.0.bn1.running_var', 'incre_modules.2.0.bn2.running_var', 'incre_modules.3.0.bn3.running_var', 'incre_modules.2.0.downsample.0.weight', 'incre_modules.3.0.downsample.1.bias', 'downsamp_modules.1.0.bias', 'incre_modules.1.0.bn2.running_mean', 'incre_modules.1.0.downsample.1.num_batches_tracked'} {'ocr_distri_head.object_context_block.f_down.1.0.bias', 'ocr_distri_head.object_context_block.f_object.1.0.num_batches_tracked', 'ocr_distri_head.conv_bn_dropout.1.0.bias', 'conv3x3_ocr.0.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.weight', 'ocr_distri_head.object_context_block.f_pixel.2.weight', 'ocr_distri_head.conv_bn_dropout.1.0.running_var', 'ocr_distri_head.object_context_block.f_pixel.3.0.running_var', 'ocr_distri_head.object_context_block.f_object.3.0.num_batches_tracked', 'ocr_distri_head.conv_bn_dropout.0.weight', 'ocr_distri_head.object_context_block.f_object.1.0.running_var', 'ocr_distri_head.object_context_block.f_up.1.0.weight', 'conv3x3_ocr.0.bias', 'aux_head.3.bias', 'ocr_distri_head.object_context_block.f_object.3.0.weight', 'cls_head.bias', 'conv3x3_ocr.1.bias', 'ocr_distri_head.object_context_block.f_up.1.0.running_mean', 'ocr_distri_head.object_context_block.f_object.2.weight', 'ocr_distri_head.object_context_block.f_object.3.0.running_var', 'aux_head.3.weight', 'ocr_distri_head.conv_bn_dropout.1.0.num_batches_tracked', 'aux_head.1.running_mean', 'ocr_distri_head.object_context_block.f_up.0.weight', 'ocr_distri_head.object_context_block.f_object.3.0.running_mean', 'ocr_distri_head.object_context_block.f_object.1.0.weight', 'ocr_distri_head.object_context_block.f_object.1.0.running_mean', 'ocr_distri_head.object_context_block.f_up.1.0.running_var', 'ocr_distri_head.object_context_block.f_pixel.3.0.num_batches_tracked', 'aux_head.0.weight', 'aux_head.1.bias', 'ocr_distri_head.object_context_block.f_pixel.1.0.running_var', 'ocr_distri_head.object_context_block.f_pixel.1.0.bias', 'ocr_distri_head.object_context_block.f_object.0.weight', 'conv3x3_ocr.1.weight', 'ocr_distri_head.conv_bn_dropout.1.0.running_mean', 'ocr_distri_head.object_context_block.f_pixel.1.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_object.1.0.bias', 'ocr_distri_head.object_context_block.f_pixel.1.0.weight', 'aux_head.1.num_batches_tracked', 'conv3x3_ocr.1.running_mean', 'conv3x3_ocr.1.num_batches_tracked', 'ocr_distri_head.object_context_block.f_down.1.0.num_batches_tracked', 'aux_head.1.weight', 'aux_head.0.bias', 'ocr_distri_head.object_context_block.f_object.3.0.bias', 'ocr_distri_head.object_context_block.f_pixel.3.0.bias', 'ocr_distri_head.object_context_block.f_pixel.1.0.running_mean', 'ocr_distri_head.object_context_block.f_down.0.weight', 'ocr_distri_head.object_context_block.f_down.1.0.weight', 'ocr_distri_head.object_context_block.f_down.1.0.running_mean', 'ocr_distri_head.object_context_block.f_down.1.0.running_var', 'ocr_distri_head.conv_bn_dropout.1.0.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.running_mean', 'ocr_distri_head.object_context_block.f_up.1.0.bias', 'ocr_distri_head.object_context_block.f_pixel.0.weight', 'aux_head.1.running_var', 'cls_head.weight', 'ocr_distri_head.object_context_block.f_up.1.0.num_batches_tracked', 'conv3x3_ocr.1.running_var'} {'incre_modules.1.0.bn2.running_var', 'incre_modules.3.0.bn3.running_var', 'incre_modules.2.0.bn3.bias', 'downsamp_modules.0.0.bias', 'incre_modules.3.0.downsample.1.running_var', 'incre_modules.0.0.bn1.weight', 'incre_modules.1.0.bn2.num_batches_tracked', 'downsamp_modules.2.1.weight', 'incre_modules.0.0.bn1.num_batches_tracked', 'incre_modules.3.0.bn3.running_mean', 'incre_modules.0.0.bn2.running_mean', 'incre_modules.3.0.bn2.running_var', 'incre_modules.1.0.bn3.bias', 'incre_modules.1.0.bn1.running_var', 'downsamp_modules.0.0.weight', 'downsamp_modules.1.1.bias', 'downsamp_modules.0.1.running_var', 'incre_modules.1.0.bn1.running_mean', 'incre_modules.3.0.downsample.1.running_mean', 'incre_modules.3.0.bn3.weight', 'final_layer.1.running_mean', 'incre_modules.1.0.downsample.1.bias', 'final_layer.0.bias', 'incre_modules.2.0.downsample.1.running_mean', 'incre_modules.1.0.conv2.weight', 'incre_modules.2.0.conv3.weight', 'incre_modules.1.0.bn1.weight', 'incre_modules.0.0.downsample.0.weight', 'incre_modules.3.0.bn1.bias', 'incre_modules.3.0.conv2.weight', 'incre_modules.3.0.bn2.weight', 'incre_modules.0.0.conv1.weight', 'incre_modules.0.0.bn3.running_var', 'incre_modules.1.0.bn3.running_mean', 'incre_modules.3.0.downsample.1.num_batches_tracked', 'incre_modules.2.0.bn1.weight', 'downsamp_modules.2.1.num_batches_tracked', 'incre_modules.0.0.bn2.num_batches_tracked', 'incre_modules.1.0.bn2.bias', 'downsamp_modules.1.1.running_mean', 'incre_modules.3.0.bn1.num_batches_tracked', 'incre_modules.0.0.conv2.weight', 'incre_modules.2.0.bn3.num_batches_tracked', 'final_layer.1.num_batches_tracked', 'incre_modules.3.0.bn2.bias', 'incre_modules.1.0.downsample.1.running_var', 'downsamp_modules.2.0.bias', 'incre_modules.2.0.bn3.running_mean', 'incre_modules.2.0.bn1.num_batches_tracked', 'incre_modules.3.0.conv1.weight', 'downsamp_modules.0.1.bias', 'classifier.weight', 'incre_modules.0.0.bn1.running_var', 'final_layer.1.running_var', 'final_layer.1.bias', 'incre_modules.0.0.bn3.num_batches_tracked', 'incre_modules.1.0.bn1.bias', 'incre_modules.0.0.bn3.bias', 'final_layer.0.weight', 'classifier.bias', 'incre_modules.2.0.conv2.weight', 'downsamp_modules.2.1.running_var', 'downsamp_modules.1.0.weight', 'incre_modules.0.0.downsample.1.num_batches_tracked', 'incre_modules.1.0.bn3.weight', 'incre_modules.2.0.downsample.1.running_var', 'incre_modules.0.0.bn3.running_mean', 'downsamp_modules.2.0.weight', 'incre_modules.2.0.bn1.bias', 'incre_modules.3.0.downsample.1.bias', 'incre_modules.1.0.downsample.1.num_batches_tracked', 'incre_modules.2.0.bn3.running_var', 'incre_modules.2.0.downsample.1.num_batches_tracked', 'incre_modules.1.0.bn1.num_batches_tracked', 'incre_modules.2.0.bn1.running_mean', 'incre_modules.2.0.conv1.weight', 'incre_modules.2.0.bn2.weight', 'incre_modules.1.0.bn3.running_var', 'incre_modules.1.0.downsample.1.weight', 'downsamp_modules.0.1.weight', 'downsamp_modules.1.0.bias', 'incre_modules.2.0.bn2.num_batches_tracked', 'incre_modules.1.0.conv1.weight', 'incre_modules.1.0.bn3.num_batches_tracked', 'incre_modules.3.0.conv3.weight', 'incre_modules.2.0.downsample.1.bias', 'incre_modules.3.0.bn1.running_mean', 'downsamp_modules.0.1.num_batches_tracked', 'incre_modules.3.0.bn1.weight', 'incre_modules.2.0.bn1.running_var', 'incre_modules.3.0.downsample.0.weight', 'incre_modules.3.0.downsample.1.weight', 'incre_modules.3.0.bn2.running_mean', 'incre_modules.2.0.bn3.weight', 'incre_modules.0.0.conv3.weight', 'incre_modules.0.0.downsample.1.bias', 'downsamp_modules.1.1.num_batches_tracked', 'incre_modules.0.0.bn3.weight', 'incre_modules.1.0.conv3.weight', 'final_layer.1.weight', 'incre_modules.2.0.bn2.bias', 'downsamp_modules.2.1.running_mean', 'incre_modules.2.0.bn2.running_mean', 'incre_modules.1.0.bn2.running_mean', 'incre_modules.3.0.bn3.num_batches_tracked', 'incre_modules.1.0.downsample.0.weight', 'incre_modules.1.0.downsample.1.running_mean', 'incre_modules.0.0.downsample.1.running_mean', 'incre_modules.0.0.bn1.running_mean', 'incre_modules.0.0.bn2.bias', 'downsamp_modules.1.1.weight', 'downsamp_modules.2.1.bias', 'incre_modules.2.0.bn2.running_var', 'incre_modules.0.0.bn2.running_var', 'incre_modules.3.0.bn3.bias', 'incre_modules.3.0.bn1.running_var', 'incre_modules.0.0.bn2.weight', 'incre_modules.2.0.downsample.0.weight', 'downsamp_modules.1.1.running_var', 'incre_modules.3.0.bn2.num_batches_tracked', 'incre_modules.0.0.bn1.bias', 'incre_modules.2.0.downsample.1.weight', 'incre_modules.1.0.bn2.weight', 'downsamp_modules.0.1.running_mean', 'incre_modules.0.0.downsample.1.weight', 'incre_modules.0.0.downsample.1.running_var'} => loading pretrained model pretrained_models/hrnetv2_w48_imagenet_pretrained.pth {'ocr_distri_head.object_context_block.f_pixel.1.0.bias', 'ocr_distri_head.object_context_block.f_pixel.3.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_pixel.3.0.running_var', 'ocr_distri_head.object_context_block.f_object.1.0.weight', 'ocr_distri_head.object_context_block.f_pixel.1.0.running_var', 'aux_head.0.weight', 'ocr_distri_head.object_context_block.f_down.1.0.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.running_mean', 'cls_head.weight', 'ocr_distri_head.object_context_block.f_down.1.0.running_mean', 'ocr_distri_head.object_context_block.f_pixel.1.0.running_mean', 'ocr_distri_head.object_context_block.f_object.3.0.running_var', 'aux_head.1.num_batches_tracked', 'aux_head.1.running_mean', 'ocr_distri_head.object_context_block.f_object.3.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_up.1.0.num_batches_tracked', 'ocr_distri_head.conv_bn_dropout.1.0.running_var', 'aux_head.1.bias', 'aux_head.1.running_var', 'ocr_distri_head.object_context_block.f_object.0.weight', 'ocr_distri_head.conv_bn_dropout.1.0.weight', 'ocr_distri_head.object_context_block.f_down.0.weight', 'conv3x3_ocr.0.weight', 'ocr_distri_head.conv_bn_dropout.1.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_pixel.2.weight', 'ocr_distri_head.object_context_block.f_object.1.0.running_var', 'ocr_distri_head.object_context_block.f_object.1.0.bias', 'conv3x3_ocr.0.bias', 'conv3x3_ocr.1.num_batches_tracked', 'aux_head.3.bias', 'ocr_distri_head.object_context_block.f_object.3.0.running_mean', 'conv3x3_ocr.1.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.weight', 'ocr_distri_head.conv_bn_dropout.1.0.running_mean', 'ocr_distri_head.object_context_block.f_up.1.0.running_mean', 'ocr_distri_head.object_context_block.f_pixel.1.0.weight', 'conv3x3_ocr.1.running_mean', 'ocr_distri_head.object_context_block.f_down.1.0.num_batches_tracked', 'aux_head.3.weight', 'ocr_distri_head.object_context_block.f_up.0.weight', 'ocr_distri_head.object_context_block.f_up.1.0.bias', 'ocr_distri_head.object_context_block.f_object.3.0.weight', 'ocr_distri_head.object_context_block.f_up.1.0.running_var', 'ocr_distri_head.object_context_block.f_down.1.0.bias', 'ocr_distri_head.object_context_block.f_pixel.1.0.num_batches_tracked', 'ocr_distri_head.conv_bn_dropout.1.0.bias', 'cls_head.bias', 'ocr_distri_head.object_context_block.f_pixel.3.0.bias', 'ocr_distri_head.object_context_block.f_down.1.0.running_var', 'conv3x3_ocr.1.running_var', 'conv3x3_ocr.1.bias', 'ocr_distri_head.object_context_block.f_object.1.0.num_batches_tracked', 'aux_head.0.bias', 'ocr_distri_head.conv_bn_dropout.0.weight', 'ocr_distri_head.object_context_block.f_object.2.weight', 'aux_head.1.weight', 'ocr_distri_head.object_context_block.f_pixel.0.weight', 'ocr_distri_head.object_context_block.f_up.1.0.weight', 'ocr_distri_head.object_context_block.f_object.1.0.running_mean', 'ocr_distri_head.object_context_block.f_object.3.0.bias'} {'incre_modules.3.0.bn1.running_mean', 'incre_modules.0.0.bn1.running_var', 'downsamp_modules.0.1.running_var', 'incre_modules.0.0.downsample.1.running_mean', 'incre_modules.0.0.bn2.bias', 'incre_modules.2.0.bn2.weight', 'incre_modules.0.0.bn1.weight', 'final_layer.1.running_var', 'downsamp_modules.1.1.bias', 'downsamp_modules.0.1.running_mean', 'incre_modules.0.0.bn3.running_var', 'incre_modules.3.0.bn2.running_mean', 'incre_modules.3.0.bn3.num_batches_tracked', 'downsamp_modules.1.0.weight', 'downsamp_modules.2.0.bias', 'incre_modules.3.0.downsample.1.weight', 'incre_modules.0.0.conv3.weight', 'downsamp_modules.1.1.weight', 'incre_modules.3.0.conv3.weight', 'incre_modules.2.0.bn3.weight', 'incre_modules.2.0.bn1.running_mean', 'incre_modules.2.0.downsample.1.bias', 'downsamp_modules.2.1.weight', 'incre_modules.1.0.bn2.bias', 'incre_modules.2.0.bn2.running_var', 'final_layer.0.weight', 'downsamp_modules.2.1.running_mean', 'incre_modules.0.0.bn3.num_batches_tracked', 'incre_modules.1.0.downsample.1.weight', 'incre_modules.0.0.downsample.1.running_var', 'incre_modules.2.0.bn1.num_batches_tracked', 'incre_modules.1.0.downsample.1.num_batches_tracked', 'classifier.weight', 'incre_modules.1.0.bn2.running_mean', 'incre_modules.0.0.conv1.weight', 'incre_modules.1.0.bn3.weight', 'downsamp_modules.2.1.running_var', 'incre_modules.3.0.bn2.weight', 'incre_modules.1.0.bn2.weight', 'incre_modules.2.0.bn3.num_batches_tracked', 'incre_modules.1.0.bn1.running_mean', 'final_layer.0.bias', 'incre_modules.0.0.bn1.num_batches_tracked', 'incre_modules.3.0.bn2.bias', 'incre_modules.1.0.bn1.running_var', 'incre_modules.0.0.downsample.0.weight', 'final_layer.1.weight', 'incre_modules.1.0.bn1.num_batches_tracked', 'incre_modules.1.0.bn3.num_batches_tracked', 'downsamp_modules.0.1.weight', 'incre_modules.2.0.conv2.weight', 'incre_modules.2.0.bn2.running_mean', 'classifier.bias', 'incre_modules.2.0.downsample.1.num_batches_tracked', 'incre_modules.0.0.bn2.running_var', 'downsamp_modules.1.1.num_batches_tracked', 'incre_modules.3.0.bn3.bias', 'incre_modules.2.0.downsample.1.running_mean', 'incre_modules.3.0.downsample.0.weight', 'incre_modules.2.0.bn3.running_var', 'downsamp_modules.0.0.weight', 'incre_modules.2.0.bn2.num_batches_tracked', 'incre_modules.1.0.conv1.weight', 'incre_modules.3.0.conv2.weight', 'incre_modules.2.0.conv1.weight', 'incre_modules.2.0.bn3.bias', 'incre_modules.3.0.downsample.1.bias', 'downsamp_modules.0.0.bias', 'incre_modules.3.0.bn1.bias', 'incre_modules.1.0.downsample.1.bias', 'incre_modules.3.0.downsample.1.running_var', 'downsamp_modules.2.0.weight', 'incre_modules.3.0.bn3.weight', 'incre_modules.3.0.bn1.weight', 'incre_modules.1.0.downsample.0.weight', 'incre_modules.1.0.downsample.1.running_var', 'incre_modules.0.0.bn2.num_batches_tracked', 'final_layer.1.bias', 'incre_modules.0.0.conv2.weight', 'incre_modules.0.0.bn3.weight', 'incre_modules.3.0.bn2.num_batches_tracked', 'final_layer.1.running_mean', 'incre_modules.2.0.bn1.bias', 'incre_modules.3.0.bn1.running_var', 'downsamp_modules.1.1.running_var', 'incre_modules.1.0.bn1.weight', 'incre_modules.3.0.bn3.running_var', 'incre_modules.1.0.conv2.weight', 'incre_modules.2.0.bn1.weight', 'incre_modules.1.0.bn2.running_var', 'downsamp_modules.2.1.num_batches_tracked', 'incre_modules.0.0.bn1.bias', 'incre_modules.3.0.bn1.num_batches_tracked', 'incre_modules.0.0.bn3.bias', 'incre_modules.1.0.bn2.num_batches_tracked', 'incre_modules.0.0.downsample.1.weight', 'incre_modules.3.0.bn3.running_mean', 'incre_modules.3.0.conv1.weight', 'incre_modules.1.0.bn3.running_var', 'incre_modules.1.0.conv3.weight', 'incre_modules.0.0.bn1.running_mean', 'incre_modules.2.0.downsample.0.weight', 'downsamp_modules.0.1.bias', 'incre_modules.2.0.bn1.running_var', 'incre_modules.3.0.downsample.1.num_batches_tracked', 'final_layer.1.num_batches_tracked', 'incre_modules.1.0.bn3.running_mean', 'incre_modules.2.0.bn2.bias', 'downsamp_modules.0.1.num_batches_tracked', 'incre_modules.0.0.downsample.1.bias', 'incre_modules.2.0.downsample.1.weight', 'downsamp_modules.2.1.bias', 'incre_modules.0.0.downsample.1.num_batches_tracked', 'incre_modules.2.0.downsample.1.running_var', 'incre_modules.0.0.bn2.weight', 'downsamp_modules.1.1.running_mean', 'incre_modules.0.0.bn2.running_mean', 'incre_modules.1.0.downsample.1.running_mean', 'incre_modules.0.0.bn3.running_mean', 'incre_modules.1.0.bn3.bias', 'downsamp_modules.1.0.bias', 'incre_modules.1.0.bn1.bias', 'incre_modules.2.0.conv3.weight', 'incre_modules.3.0.downsample.1.running_mean', 'incre_modules.2.0.bn3.running_mean', 'incre_modules.3.0.bn2.running_var'} => loading pretrained model pretrained_models/hrnetv2_w48_imagenet_pretrained.pth {'ocr_distri_head.object_context_block.f_up.1.0.running_mean', 'conv3x3_ocr.0.weight', 'cls_head.bias', 'ocr_distri_head.conv_bn_dropout.1.0.running_mean', 'ocr_distri_head.object_context_block.f_object.1.0.weight', 'ocr_distri_head.object_context_block.f_object.3.0.running_mean', 'ocr_distri_head.conv_bn_dropout.1.0.weight', 'ocr_distri_head.object_context_block.f_up.1.0.num_batches_tracked', 'aux_head.1.num_batches_tracked', 'ocr_distri_head.conv_bn_dropout.0.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.bias', 'ocr_distri_head.object_context_block.f_pixel.3.0.weight', 'ocr_distri_head.object_context_block.f_pixel.3.0.running_var', 'ocr_distri_head.object_context_block.f_pixel.3.0.running_mean', 'ocr_distri_head.object_context_block.f_up.0.weight', 'aux_head.3.weight', 'conv3x3_ocr.1.running_mean', 'aux_head.1.running_var', 'ocr_distri_head.object_context_block.f_down.0.weight', 'ocr_distri_head.object_context_block.f_object.3.0.running_var', 'ocr_distri_head.conv_bn_dropout.1.0.bias', 'ocr_distri_head.object_context_block.f_pixel.0.weight', 'ocr_distri_head.object_context_block.f_up.1.0.bias', 'ocr_distri_head.object_context_block.f_object.0.weight', 'aux_head.0.bias', 'ocr_distri_head.object_context_block.f_down.1.0.running_mean', 'ocr_distri_head.object_context_block.f_object.3.0.bias', 'ocr_distri_head.object_context_block.f_object.2.weight', 'ocr_distri_head.object_context_block.f_pixel.1.0.running_var', 'conv3x3_ocr.1.weight', 'ocr_distri_head.conv_bn_dropout.1.0.running_var', 'aux_head.3.bias', 'conv3x3_ocr.1.bias', 'ocr_distri_head.object_context_block.f_pixel.2.weight', 'ocr_distri_head.object_context_block.f_down.1.0.running_var', 'ocr_distri_head.object_context_block.f_object.3.0.num_batches_tracked', 'cls_head.weight', 'ocr_distri_head.object_context_block.f_down.1.0.bias', 'ocr_distri_head.conv_bn_dropout.1.0.num_batches_tracked', 'aux_head.1.running_mean', 'ocr_distri_head.object_context_block.f_up.1.0.running_var', 'conv3x3_ocr.1.running_var', 'conv3x3_ocr.0.bias', 'aux_head.1.weight', 'ocr_distri_head.object_context_block.f_pixel.1.0.weight', 'aux_head.0.weight', 'ocr_distri_head.object_context_block.f_object.1.0.running_mean', 'ocr_distri_head.object_context_block.f_up.1.0.weight', 'aux_head.1.bias', 'ocr_distri_head.object_context_block.f_pixel.1.0.running_mean', 'conv3x3_ocr.1.num_batches_tracked', 'ocr_distri_head.object_context_block.f_down.1.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_pixel.1.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_pixel.3.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_object.1.0.num_batches_tracked', 'ocr_distri_head.object_context_block.f_object.1.0.running_var', 'ocr_distri_head.object_context_block.f_pixel.1.0.bias', 'ocr_distri_head.object_context_block.f_down.1.0.weight', 'ocr_distri_head.object_context_block.f_object.3.0.weight', 'ocr_distri_head.object_context_block.f_object.1.0.bias'} {'incre_modules.3.0.bn2.weight', 'downsamp_modules.1.1.bias', 'incre_modules.0.0.bn3.num_batches_tracked', 'incre_modules.2.0.bn3.running_mean', 'incre_modules.2.0.downsample.1.weight', 'incre_modules.2.0.conv3.weight', 'incre_modules.0.0.downsample.1.running_var', 'incre_modules.0.0.bn1.running_mean', 'final_layer.1.bias', 'incre_modules.1.0.bn1.running_mean', 'downsamp_modules.2.1.weight', 'incre_modules.1.0.bn3.num_batches_tracked', 'incre_modules.0.0.conv1.weight', 'incre_modules.1.0.downsample.1.num_batches_tracked', 'downsamp_modules.2.1.running_var', 'incre_modules.3.0.bn2.bias', 'incre_modules.3.0.downsample.1.weight', 'incre_modules.0.0.downsample.1.bias', 'incre_modules.2.0.bn1.bias', 'downsamp_modules.0.0.bias', 'downsamp_modules.0.0.weight', 'incre_modules.1.0.bn3.weight', 'incre_modules.2.0.downsample.1.running_mean', 'incre_modules.0.0.bn3.bias', 'downsamp_modules.0.1.running_var', 'incre_modules.0.0.bn2.weight', 'incre_modules.0.0.bn3.running_mean', 'incre_modules.0.0.bn1.num_batches_tracked', 'incre_modules.0.0.bn2.running_mean', 'incre_modules.0.0.conv2.weight', 'incre_modules.3.0.conv2.weight', 'incre_modules.0.0.bn2.bias', 'final_layer.1.weight', 'incre_modules.2.0.bn1.running_mean', 'incre_modules.0.0.bn1.weight', 'incre_modules.0.0.bn1.running_var', 'classifier.bias', 'incre_modules.2.0.bn1.num_batches_tracked', 'incre_modules.1.0.bn2.running_var', 'incre_modules.1.0.downsample.0.weight', 'incre_modules.1.0.bn3.running_var', 'downsamp_modules.1.1.running_var', 'incre_modules.3.0.downsample.0.weight', 'incre_modules.2.0.downsample.0.weight', 'incre_modules.3.0.downsample.1.num_batches_tracked', 'final_layer.1.running_mean', 'incre_modules.2.0.bn1.running_var', 'downsamp_modules.0.1.bias', 'incre_modules.1.0.bn2.running_mean', 'incre_modules.2.0.downsample.1.num_batches_tracked', 'downsamp_modules.2.1.num_batches_tracked', 'incre_modules.0.0.bn3.weight', 'incre_modules.0.0.bn2.running_var', 'incre_modules.1.0.bn1.weight', 'incre_modules.0.0.bn2.num_batches_tracked', 'incre_modules.2.0.bn3.bias', 'incre_modules.2.0.bn3.num_batches_tracked', 'incre_modules.3.0.downsample.1.running_var', 'incre_modules.1.0.conv2.weight', 'final_layer.1.running_var', 'incre_modules.3.0.bn2.running_var', 'incre_modules.3.0.conv3.weight', 'incre_modules.1.0.bn2.num_batches_tracked', 'incre_modules.0.0.bn1.bias', 'incre_modules.3.0.downsample.1.bias', 'incre_modules.1.0.downsample.1.running_var', 'incre_modules.3.0.bn3.weight', 'incre_modules.1.0.bn3.running_mean', 'incre_modules.1.0.bn1.num_batches_tracked', 'downsamp_modules.1.0.weight', 'incre_modules.3.0.bn1.running_var', 'incre_modules.1.0.bn1.bias', 'incre_modules.2.0.bn2.bias', 'incre_modules.0.0.downsample.1.weight', 'incre_modules.2.0.conv1.weight', 'incre_modules.2.0.conv2.weight', 'incre_modules.3.0.bn3.running_mean', 'incre_modules.0.0.downsample.1.num_batches_tracked', 'classifier.weight', 'incre_modules.2.0.downsample.1.bias', 'incre_modules.0.0.bn3.running_var', 'incre_modules.3.0.bn2.running_mean', 'incre_modules.0.0.downsample.0.weight', 'incre_modules.1.0.downsample.1.weight', 'downsamp_modules.1.0.bias', 'incre_modules.2.0.bn1.weight', 'incre_modules.0.0.conv3.weight', 'incre_modules.1.0.conv1.weight', 'incre_modules.3.0.downsample.1.running_mean', 'downsamp_modules.1.1.running_mean', 'incre_modules.1.0.bn2.bias', 'incre_modules.2.0.bn2.running_var', 'incre_modules.1.0.bn3.bias', 'incre_modules.2.0.downsample.1.running_var', 'incre_modules.2.0.bn3.weight', 'incre_modules.3.0.bn1.running_mean', 'downsamp_modules.0.1.running_mean', 'final_layer.1.num_batches_tracked', 'incre_modules.2.0.bn2.num_batches_tracked', 'final_layer.0.weight', 'incre_modules.3.0.conv1.weight', 'incre_modules.3.0.bn1.weight', 'incre_modules.3.0.bn2.num_batches_tracked', 'incre_modules.2.0.bn3.running_var', 'downsamp_modules.2.0.weight', 'incre_modules.0.0.downsample.1.running_mean', 'incre_modules.2.0.bn2.weight', 'incre_modules.2.0.bn2.running_mean', 'incre_modules.3.0.bn3.running_var', 'incre_modules.1.0.downsample.1.running_mean', 'downsamp_modules.2.1.bias', 'downsamp_modules.0.1.num_batches_tracked', 'downsamp_modules.1.1.num_batches_tracked', 'downsamp_modules.2.0.bias', 'incre_modules.3.0.bn1.num_batches_tracked', 'downsamp_modules.2.1.running_mean', 'incre_modules.3.0.bn1.bias', 'incre_modules.1.0.bn1.running_var', 'downsamp_modules.1.1.weight', 'incre_modules.3.0.bn3.bias', 'incre_modules.1.0.downsample.1.bias', 'incre_modules.3.0.bn3.num_batches_tracked', 'incre_modules.1.0.bn2.weight', 'incre_modules.1.0.conv3.weight', 'downsamp_modules.0.1.weight', 'final_layer.0.bias'}

=> saving checkpoint to output/cityscapes/seg_hrnet_ocr_w48_train_512x1024_sgd_lr1e-2_wd5e-4_bs_12_epoch484checkpoint.pth.tar free(): invalid pointer Loss: 0.158, MeanIU: 0.8084, Best_mIoU: 0.8099 [0.98559111 0.87849699 0.93522475 0.59377196 0.65188462 0.71243647 0.75491625 0.83216277 0.93221823 0.64999572 0.95340937 0.84584251 0.6583003 0.95840466 0.83742979 0.90874941 0.77005884 0.70262829 0.79746064] Hours: 52 Done

NikhilChowdary-MCW commented 3 years ago

Hi @purse1996 ,

1) Have you made any changes in the validation dataset after training? 2) Are you using correct .PTH file for validation? Always try to validate using best.pth for better results.