KeyError: 'bottleneck.weight'

[aliencaocao]

在尝试用自己的数据finetune，加载了MSMT17的预训练模型。用的指令如下

python train.py --config_file TIL.yml MODEL.PRETRAIN_CHOICE 'self' MODEL.PRETRAIN_PATH 'swin_base_msmt17.pth' OUTPUT_DIR './log' SOLVER.BASE_LR 0.0002 SOLVER.OPTIMIZER_NAME 'SGD' MODEL.SEMANTIC_WEIGHT 0.2

yml文件

MODEL:
  PRETRAIN_HW_RATIO: 2
  METRIC_LOSS_TYPE: 'triplet'
  IF_LABELSMOOTH: 'off'
  IF_WITH_CENTER: 'no'
  NAME: 'transformer'
  NO_MARGIN: True
  DEVICE_ID: ('0')
  TRANSFORMER_TYPE: 'swin_base_patch4_window7_224'
  STRIDE_SIZE: [16, 16]

INPUT:
  SIZE_TRAIN: [384, 128]
  SIZE_TEST: [384, 128]
  PROB: 0.5 # random horizontal flip
  RE_PROB: 0.5 # random erasing
  PADDING: 10
  PIXEL_MEAN: [0.5, 0.5, 0.5]
  PIXEL_STD: [0.5, 0.5, 0.5]

DATASETS:
  NAMES: ('TIL')
  ROOT_DIR: ('../cv_sample_data_yolo/reid')

DATALOADER:
  SAMPLER: 'softmax_triplet'
  NUM_INSTANCE: 2
  NUM_WORKERS: 4

SOLVER:
  OPTIMIZER_NAME: 'SGD'
  MAX_EPOCHS: 120
  BASE_LR: 0.0008
  WARMUP_EPOCHS: 20
  IMS_PER_BATCH: 64
  WARMUP_METHOD: 'cosine'
  LARGE_FC_LR: False
  CHECKPOINT_PERIOD: 120
  LOG_PERIOD: 20
  EVAL_PERIOD: 10
  WEIGHT_DECAY:  1e-4
  WEIGHT_DECAY_BIAS: 1e-4
  BIAS_LR_FACTOR: 2

TEST:
  EVAL: True
  IMS_PER_BATCH: 256
  RE_RANKING: True
  WEIGHT: ''
  NECK_FEAT: 'before'
  FEAT_NORM: 'yes'

OUTPUT_DIR: './log/'

已经根据#5 (comment) 替换了代码

Traceback (most recent call last):
  File "C:\Users\alien\Documents\PyCharm-Projects\TIL-2023\CV\SOLIDER-REID\train.py", line 71, in <module>
    model = make_model(cfg, num_class=num_classes, camera_num=camera_num, view_num = view_num, semantic_weight = cfg.MODEL.SEMANTIC_WEIGHT)
  File "C:\Users\alien\Documents\PyCharm-Projects\TIL-2023\CV\SOLIDER-REID\model\make_model.py", line 449, in make_model
    model = build_transformer(num_class, camera_num, view_num, cfg, __factory_T_type, semantic_weight)
  File "C:\Users\alien\Documents\PyCharm-Projects\TIL-2023\CV\SOLIDER-REID\model\make_model.py", line 198, in __init__
    self.state_dict()[i].copy_(param_dict[i])
KeyError: 'bottleneck.weight

[cwhgn]

您好，方便把您修改后的self.state_dict()[i].copy_(param_dict[i])所在函数的代码贴一下么？

[aliencaocao]

make model.py line 190

        convert_weights = True if pretrain_choice == 'imagenet' else False
        self.base = factory[cfg.MODEL.TRANSFORMER_TYPE](img_size=cfg.INPUT.SIZE_TRAIN, drop_path_rate=cfg.MODEL.DROP_PATH, drop_rate= cfg.MODEL.DROP_OUT,attn_drop_rate=cfg.MODEL.ATT_DROP_RATE, pretrained=model_path, convert_weights=convert_weights, semantic_weight=semantic_weight)
        if model_path != '':
            print(self.state_dict().keys())
            param_dict = torch.load(model_path)
            for i in param_dict:
                if 'classifier' in i:
                    continue
                self.state_dict()[i].copy_(param_dict[i])
        self.in_planes = self.base.num_features[-1]

        self.num_classes = num_classes
        self.ID_LOSS_TYPE = cfg.MODEL.ID_LOSS_TYPE

[aliencaocao]

odict_keys(['base.patch_embed.projection.weight', 'base.patch_embed.projection.bias', 'base.patch_embed.norm.weight', 'base.patch_embed.norm.bias', 'base.stages.0.blocks.0.norm1.weight', 'base.stages.0.blocks.0.norm1.bias', 'base.stages.0.blocks.0.attn.w_msa.relative_position_bias_table', 'base.stages.0.blocks.0.attn.w_msa.relative_position_index', 'base.stages.0.blocks.0.attn.w_msa.qkv.weight', 'base.stages.0.blocks.0.attn.w_msa.qkv.bias', 'base.stages.0.blocks.0.attn.w_msa.proj.weight', 'base.stages.0.blocks.0.attn.w_msa.proj.bias', 'base.stages.0.blocks.0.norm2.weight', 'base.stages.0.blocks.0.norm2.bias', 'base.stages.0.blocks.0.ffn.layers.0.0.weight', 'base.stages.0.blocks.0.ffn.layers.0.0.bias', 'base.stages.0.blocks.0.ffn.layers.1.weight', 'base.stages.0.blocks.0.ffn.layers.1.bias', 'base.stages.0.blocks.1.norm1.weight', 'base.stages.0.blocks.1.norm1.bias', 'base.stages.0.blocks.1.attn.w_msa.relative_position_bias_table', 'base.stages.0.blocks.1.attn.w_msa.relative_position_index', 'base.stages.0.blocks.1.attn.w_msa.qkv.weight', 'base.stages.0.blocks.1.attn.w_msa.qkv.bias', 'base.stages.0.blocks.1.attn.w_msa.proj.weight', 'base.stages.0.blocks.1.attn.w_msa.proj.bias', 'base.stages.0.blocks.1.norm2.weight', 'base.stages.0.blocks.1.norm2.bias', 'base.stages.0.blocks.1.ffn.layers.0.0.weight', 'base.stages.0.blocks.1.ffn.layers.0.0.bias', 'base.stages.0.blocks.1.ffn.layers.1.weight', 'base.stages.0.blocks.1.ffn.layers.1.bias', 'base.stages.0.downsample.norm.weight', 'base.stages.0.downsample.norm.bias', 'base.stages.0.downsample.reduction.weight', 'base.stages.1.blocks.0.norm1.weight', 'base.stages.1.blocks.0.norm1.bias', 'base.stages.1.blocks.0.attn.w_msa.relative_position_bias_table', 'base.stages.1.blocks.0.attn.w_msa.relative_position_index', 'base.stages.1.blocks.0.attn.w_msa.qkv.weight', 'base.stages.1.blocks.0.attn.w_msa.qkv.bias', 'base.stages.1.blocks.0.attn.w_msa.proj.weight', 'base.stages.1.blocks.0.attn.w_msa.proj.bias', 'base.stages.1.blocks.0.norm2.weight', 'base.stages.1.blocks.0.norm2.bias', 'base.stages.1.blocks.0.ffn.layers.0.0.weight', 'base.stages.1.blocks.0.ffn.layers.0.0.bias', 'base.stages.1.blocks.0.ffn.layers.1.weight', 'base.stages.1.blocks.0.ffn.layers.1.bias', 'base.stages.1.blocks.1.norm1.weight', 'base.stages.1.blocks.1.norm1.bias', 'base.stages.1.blocks.1.attn.w_msa.relative_position_bias_table', 'base.stages.1.blocks.1.attn.w_msa.relative_position_index', 'base.stages.1.blocks.1.attn.w_msa.qkv.weight', 'base.stages.1.blocks.1.attn.w_msa.qkv.bias', 'base.stages.1.blocks.1.attn.w_msa.proj.weight', 'base.stages.1.blocks.1.attn.w_msa.proj.bias', 'base.stages.1.blocks.1.norm2.weight', 'base.stages.1.blocks.1.norm2.bias', 'base.stages.1.blocks.1.ffn.layers.0.0.weight', 'base.stages.1.blocks.1.ffn.layers.0.0.bias', 'base.stages.1.blocks.1.ffn.layers.1.weight', 'base.stages.1.blocks.1.ffn.layers.1.bias', 'base.stages.1.downsample.norm.weight', 'base.stages.1.downsample.norm.bias', 'base.stages.1.downsample.reduction.weight', 'base.stages.2.blocks.0.norm1.weight', 'base.stages.2.blocks.0.norm1.bias', 'base.stages.2.blocks.0.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.0.attn.w_msa.relative_position_index', 'base.stages.2.blocks.0.attn.w_msa.qkv.weight', 'base.stages.2.blocks.0.attn.w_msa.qkv.bias', 'base.stages.2.blocks.0.attn.w_msa.proj.weight', 'base.stages.2.blocks.0.attn.w_msa.proj.bias', 'base.stages.2.blocks.0.norm2.weight', 'base.stages.2.blocks.0.norm2.bias', 'base.stages.2.blocks.0.ffn.layers.0.0.weight', 'base.stages.2.blocks.0.ffn.layers.0.0.bias', 'base.stages.2.blocks.0.ffn.layers.1.weight', 'base.stages.2.blocks.0.ffn.layers.1.bias', 'base.stages.2.blocks.1.norm1.weight', 'base.stages.2.blocks.1.norm1.bias', 'base.stages.2.blocks.1.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.1.attn.w_msa.relative_position_index', 'base.stages.2.blocks.1.attn.w_msa.qkv.weight', 'base.stages.2.blocks.1.attn.w_msa.qkv.bias', 'base.stages.2.blocks.1.attn.w_msa.proj.weight', 'base.stages.2.blocks.1.attn.w_msa.proj.bias', 'base.stages.2.blocks.1.norm2.weight', 'base.stages.2.blocks.1.norm2.bias', 'base.stages.2.blocks.1.ffn.layers.0.0.weight', 'base.stages.2.blocks.1.ffn.layers.0.0.bias', 'base.stages.2.blocks.1.ffn.layers.1.weight', 'base.stages.2.blocks.1.ffn.layers.1.bias', 'base.stages.2.blocks.2.norm1.weight', 'base.stages.2.blocks.2.norm1.bias', 'base.stages.2.blocks.2.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.2.attn.w_msa.relative_position_index', 'base.stages.2.blocks.2.attn.w_msa.qkv.weight', 'base.stages.2.blocks.2.attn.w_msa.qkv.bias', 'base.stages.2.blocks.2.attn.w_msa.proj.weight', 'base.stages.2.blocks.2.attn.w_msa.proj.bias', 'base.stages.2.blocks.2.norm2.weight', 'base.stages.2.blocks.2.norm2.bias', 'base.stages.2.blocks.2.ffn.layers.0.0.weight', 'base.stages.2.blocks.2.ffn.layers.0.0.bias', 'base.stages.2.blocks.2.ffn.layers.1.weight', 'base.stages.2.blocks.2.ffn.layers.1.bias', 'base.stages.2.blocks.3.norm1.weight', 'base.stages.2.blocks.3.norm1.bias', 'base.stages.2.blocks.3.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.3.attn.w_msa.relative_position_index', 'base.stages.2.blocks.3.attn.w_msa.qkv.weight', 'base.stages.2.blocks.3.attn.w_msa.qkv.bias', 'base.stages.2.blocks.3.attn.w_msa.proj.weight', 'base.stages.2.blocks.3.attn.w_msa.proj.bias', 'base.stages.2.blocks.3.norm2.weight', 'base.stages.2.blocks.3.norm2.bias', 'base.stages.2.blocks.3.ffn.layers.0.0.weight', 'base.stages.2.blocks.3.ffn.layers.0.0.bias', 'base.stages.2.blocks.3.ffn.layers.1.weight', 'base.stages.2.blocks.3.ffn.layers.1.bias', 'base.stages.2.blocks.4.norm1.weight', 'base.stages.2.blocks.4.norm1.bias', 'base.stages.2.blocks.4.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.4.attn.w_msa.relative_position_index', 'base.stages.2.blocks.4.attn.w_msa.qkv.weight', 'base.stages.2.blocks.4.attn.w_msa.qkv.bias', 'base.stages.2.blocks.4.attn.w_msa.proj.weight', 'base.stages.2.blocks.4.attn.w_msa.proj.bias', 'base.stages.2.blocks.4.norm2.weight', 'base.stages.2.blocks.4.norm2.bias', 'base.stages.2.blocks.4.ffn.layers.0.0.weight', 'base.stages.2.blocks.4.ffn.layers.0.0.bias', 'base.stages.2.blocks.4.ffn.layers.1.weight', 'base.stages.2.blocks.4.ffn.layers.1.bias', 'base.stages.2.blocks.5.norm1.weight', 'base.stages.2.blocks.5.norm1.bias', 'base.stages.2.blocks.5.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.5.attn.w_msa.relative_position_index', 'base.stages.2.blocks.5.attn.w_msa.qkv.weight', 'base.stages.2.blocks.5.attn.w_msa.qkv.bias', 'base.stages.2.blocks.5.attn.w_msa.proj.weight', 'base.stages.2.blocks.5.attn.w_msa.proj.bias', 'base.stages.2.blocks.5.norm2.weight', 'base.stages.2.blocks.5.norm2.bias', 'base.stages.2.blocks.5.ffn.layers.0.0.weight', 'base.stages.2.blocks.5.ffn.layers.0.0.bias', 'base.stages.2.blocks.5.ffn.layers.1.weight', 'base.stages.2.blocks.5.ffn.layers.1.bias', 'base.stages.2.blocks.6.norm1.weight', 'base.stages.2.blocks.6.norm1.bias', 'base.stages.2.blocks.6.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.6.attn.w_msa.relative_position_index', 'base.stages.2.blocks.6.attn.w_msa.qkv.weight', 'base.stages.2.blocks.6.attn.w_msa.qkv.bias', 'base.stages.2.blocks.6.attn.w_msa.proj.weight', 'base.stages.2.blocks.6.attn.w_msa.proj.bias', 'base.stages.2.blocks.6.norm2.weight', 'base.stages.2.blocks.6.norm2.bias', 'base.stages.2.blocks.6.ffn.layers.0.0.weight', 'base.stages.2.blocks.6.ffn.layers.0.0.bias', 'base.stages.2.blocks.6.ffn.layers.1.weight', 'base.stages.2.blocks.6.ffn.layers.1.bias', 'base.stages.2.blocks.7.norm1.weight', 'base.stages.2.blocks.7.norm1.bias', 'base.stages.2.blocks.7.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.7.attn.w_msa.relative_position_index', 'base.stages.2.blocks.7.attn.w_msa.qkv.weight', 'base.stages.2.blocks.7.attn.w_msa.qkv.bias', 'base.stages.2.blocks.7.attn.w_msa.proj.weight', 'base.stages.2.blocks.7.attn.w_msa.proj.bias', 'base.stages.2.blocks.7.norm2.weight', 'base.stages.2.blocks.7.norm2.bias', 'base.stages.2.blocks.7.ffn.layers.0.0.weight', 'base.stages.2.blocks.7.ffn.layers.0.0.bias', 'base.stages.2.blocks.7.ffn.layers.1.weight', 'base.stages.2.blocks.7.ffn.layers.1.bias', 'base.stages.2.blocks.8.norm1.weight', 'base.stages.2.blocks.8.norm1.bias', 'base.stages.2.blocks.8.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.8.attn.w_msa.relative_position_index', 'base.stages.2.blocks.8.attn.w_msa.qkv.weight', 'base.stages.2.blocks.8.attn.w_msa.qkv.bias', 'base.stages.2.blocks.8.attn.w_msa.proj.weight', 'base.stages.2.blocks.8.attn.w_msa.proj.bias', 'base.stages.2.blocks.8.norm2.weight', 'base.stages.2.blocks.8.norm2.bias', 'base.stages.2.blocks.8.ffn.layers.0.0.weight', 'base.stages.2.blocks.8.ffn.layers.0.0.bias', 'base.stages.2.blocks.8.ffn.layers.1.weight', 'base.stages.2.blocks.8.ffn.layers.1.bias', 'base.stages.2.blocks.9.norm1.weight', 'base.stages.2.blocks.9.norm1.bias', 'base.stages.2.blocks.9.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.9.attn.w_msa.relative_position_index', 'base.stages.2.blocks.9.attn.w_msa.qkv.weight', 'base.stages.2.blocks.9.attn.w_msa.qkv.bias', 'base.stages.2.blocks.9.attn.w_msa.proj.weight', 'base.stages.2.blocks.9.attn.w_msa.proj.bias', 'base.stages.2.blocks.9.norm2.weight', 'base.stages.2.blocks.9.norm2.bias', 'base.stages.2.blocks.9.ffn.layers.0.0.weight', 'base.stages.2.blocks.9.ffn.layers.0.0.bias', 'base.stages.2.blocks.9.ffn.layers.1.weight', 'base.stages.2.blocks.9.ffn.layers.1.bias', 'base.stages.2.blocks.10.norm1.weight', 'base.stages.2.blocks.10.norm1.bias', 'base.stages.2.blocks.10.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.10.attn.w_msa.relative_position_index', 'base.stages.2.blocks.10.attn.w_msa.qkv.weight', 'base.stages.2.blocks.10.attn.w_msa.qkv.bias', 'base.stages.2.blocks.10.attn.w_msa.proj.weight', 'base.stages.2.blocks.10.attn.w_msa.proj.bias', 'base.stages.2.blocks.10.norm2.weight', 'base.stages.2.blocks.10.norm2.bias', 'base.stages.2.blocks.10.ffn.layers.0.0.weight', 'base.stages.2.blocks.10.ffn.layers.0.0.bias', 'base.stages.2.blocks.10.ffn.layers.1.weight', 'base.stages.2.blocks.10.ffn.layers.1.bias', 'base.stages.2.blocks.11.norm1.weight', 'base.stages.2.blocks.11.norm1.bias', 'base.stages.2.blocks.11.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.11.attn.w_msa.relative_position_index', 'base.stages.2.blocks.11.attn.w_msa.qkv.weight', 'base.stages.2.blocks.11.attn.w_msa.qkv.bias', 'base.stages.2.blocks.11.attn.w_msa.proj.weight', 'base.stages.2.blocks.11.attn.w_msa.proj.bias', 'base.stages.2.blocks.11.norm2.weight', 'base.stages.2.blocks.11.norm2.bias', 'base.stages.2.blocks.11.ffn.layers.0.0.weight', 'base.stages.2.blocks.11.ffn.layers.0.0.bias', 'base.stages.2.blocks.11.ffn.layers.1.weight', 'base.stages.2.blocks.11.ffn.layers.1.bias', 'base.stages.2.blocks.12.norm1.weight', 'base.stages.2.blocks.12.norm1.bias', 'base.stages.2.blocks.12.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.12.attn.w_msa.relative_position_index', 'base.stages.2.blocks.12.attn.w_msa.qkv.weight', 'base.stages.2.blocks.12.attn.w_msa.qkv.bias', 'base.stages.2.blocks.12.attn.w_msa.proj.weight', 'base.stages.2.blocks.12.attn.w_msa.proj.bias', 'base.stages.2.blocks.12.norm2.weight', 'base.stages.2.blocks.12.norm2.bias', 'base.stages.2.blocks.12.ffn.layers.0.0.weight', 'base.stages.2.blocks.12.ffn.layers.0.0.bias', 'base.stages.2.blocks.12.ffn.layers.1.weight', 'base.stages.2.blocks.12.ffn.layers.1.bias', 'base.stages.2.blocks.13.norm1.weight', 'base.stages.2.blocks.13.norm1.bias', 'base.stages.2.blocks.13.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.13.attn.w_msa.relative_position_index', 'base.stages.2.blocks.13.attn.w_msa.qkv.weight', 'base.stages.2.blocks.13.attn.w_msa.qkv.bias', 'base.stages.2.blocks.13.attn.w_msa.proj.weight', 'base.stages.2.blocks.13.attn.w_msa.proj.bias', 'base.stages.2.blocks.13.norm2.weight', 'base.stages.2.blocks.13.norm2.bias', 'base.stages.2.blocks.13.ffn.layers.0.0.weight', 'base.stages.2.blocks.13.ffn.layers.0.0.bias', 'base.stages.2.blocks.13.ffn.layers.1.weight', 'base.stages.2.blocks.13.ffn.layers.1.bias', 'base.stages.2.blocks.14.norm1.weight', 'base.stages.2.blocks.14.norm1.bias', 'base.stages.2.blocks.14.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.14.attn.w_msa.relative_position_index', 'base.stages.2.blocks.14.attn.w_msa.qkv.weight', 'base.stages.2.blocks.14.attn.w_msa.qkv.bias', 'base.stages.2.blocks.14.attn.w_msa.proj.weight', 'base.stages.2.blocks.14.attn.w_msa.proj.bias', 'base.stages.2.blocks.14.norm2.weight', 'base.stages.2.blocks.14.norm2.bias', 'base.stages.2.blocks.14.ffn.layers.0.0.weight', 'base.stages.2.blocks.14.ffn.layers.0.0.bias', 'base.stages.2.blocks.14.ffn.layers.1.weight', 'base.stages.2.blocks.14.ffn.layers.1.bias', 'base.stages.2.blocks.15.norm1.weight', 'base.stages.2.blocks.15.norm1.bias', 'base.stages.2.blocks.15.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.15.attn.w_msa.relative_position_index', 'base.stages.2.blocks.15.attn.w_msa.qkv.weight', 'base.stages.2.blocks.15.attn.w_msa.qkv.bias', 'base.stages.2.blocks.15.attn.w_msa.proj.weight', 'base.stages.2.blocks.15.attn.w_msa.proj.bias', 'base.stages.2.blocks.15.norm2.weight', 'base.stages.2.blocks.15.norm2.bias', 'base.stages.2.blocks.15.ffn.layers.0.0.weight', 'base.stages.2.blocks.15.ffn.layers.0.0.bias', 'base.stages.2.blocks.15.ffn.layers.1.weight', 'base.stages.2.blocks.15.ffn.layers.1.bias', 'base.stages.2.blocks.16.norm1.weight', 'base.stages.2.blocks.16.norm1.bias', 'base.stages.2.blocks.16.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.16.attn.w_msa.relative_position_index', 'base.stages.2.blocks.16.attn.w_msa.qkv.weight', 'base.stages.2.blocks.16.attn.w_msa.qkv.bias', 'base.stages.2.blocks.16.attn.w_msa.proj.weight', 'base.stages.2.blocks.16.attn.w_msa.proj.bias', 'base.stages.2.blocks.16.norm2.weight', 'base.stages.2.blocks.16.norm2.bias', 'base.stages.2.blocks.16.ffn.layers.0.0.weight', 'base.stages.2.blocks.16.ffn.layers.0.0.bias', 'base.stages.2.blocks.16.ffn.layers.1.weight', 'base.stages.2.blocks.16.ffn.layers.1.bias', 'base.stages.2.blocks.17.norm1.weight', 'base.stages.2.blocks.17.norm1.bias', 'base.stages.2.blocks.17.attn.w_msa.relative_position_bias_table', 'base.stages.2.blocks.17.attn.w_msa.relative_position_index', 'base.stages.2.blocks.17.attn.w_msa.qkv.weight', 'base.stages.2.blocks.17.attn.w_msa.qkv.bias', 'base.stages.2.blocks.17.attn.w_msa.proj.weight', 'base.stages.2.blocks.17.attn.w_msa.proj.bias', 'base.stages.2.blocks.17.norm2.weight', 'base.stages.2.blocks.17.norm2.bias', 'base.stages.2.blocks.17.ffn.layers.0.0.weight', 'base.stages.2.blocks.17.ffn.layers.0.0.bias', 'base.stages.2.blocks.17.ffn.layers.1.weight', 'base.stages.2.blocks.17.ffn.layers.1.bias', 'base.stages.2.downsample.norm.weight', 'base.stages.2.downsample.norm.bias', 'base.stages.2.downsample.reduction.weight', 'base.stages.3.blocks.0.norm1.weight', 'base.stages.3.blocks.0.norm1.bias', 'base.stages.3.blocks.0.attn.w_msa.relative_position_bias_table', 'base.stages.3.blocks.0.attn.w_msa.relative_position_index', 'base.stages.3.blocks.0.attn.w_msa.qkv.weight', 'base.stages.3.blocks.0.attn.w_msa.qkv.bias', 'base.stages.3.blocks.0.attn.w_msa.proj.weight', 'base.stages.3.blocks.0.attn.w_msa.proj.bias', 'base.stages.3.blocks.0.norm2.weight', 'base.stages.3.blocks.0.norm2.bias', 'base.stages.3.blocks.0.ffn.layers.0.0.weight', 'base.stages.3.blocks.0.ffn.layers.0.0.bias', 'base.stages.3.blocks.0.ffn.layers.1.weight', 'base.stages.3.blocks.0.ffn.layers.1.bias', 'base.stages.3.blocks.1.norm1.weight', 'base.stages.3.blocks.1.norm1.bias', 'base.stages.3.blocks.1.attn.w_msa.relative_position_bias_table', 'base.stages.3.blocks.1.attn.w_msa.relative_position_index', 'base.stages.3.blocks.1.attn.w_msa.qkv.weight', 'base.stages.3.blocks.1.attn.w_msa.qkv.bias', 'base.stages.3.blocks.1.attn.w_msa.proj.weight', 'base.stages.3.blocks.1.attn.w_msa.proj.bias', 'base.stages.3.blocks.1.norm2.weight', 'base.stages.3.blocks.1.norm2.bias', 'base.stages.3.blocks.1.ffn.layers.0.0.weight', 'base.stages.3.blocks.1.ffn.layers.0.0.bias', 'base.stages.3.blocks.1.ffn.layers.1.weight', 'base.stages.3.blocks.1.ffn.layers.1.bias', 'base.norm0.weight', 'base.norm0.bias', 'base.norm1.weight', 'base.norm1.bias', 'base.norm2.weight', 'base.norm2.bias', 'base.norm3.weight', 'base.norm3.bias', 'base.semantic_embed_w.0.weight', 'base.semantic_embed_w.0.bias', 'base.semantic_embed_w.1.weight', 'base.semantic_embed_w.1.bias', 'base.semantic_embed_w.2.weight', 'base.semantic_embed_w.2.bias', 'base.semantic_embed_w.3.weight', 'base.semantic_embed_w.3.bias', 'base.semantic_embed_b.0.weight', 'base.semantic_embed_b.0.bias', 'base.semantic_embed_b.1.weight', 'base.semantic_embed_b.1.bias', 'base.semantic_embed_b.2.weight', 'base.semantic_embed_b.2.bias', 'base.semantic_embed_b.3.weight', 'base.semantic_embed_b.3.bias'])

这个是self.state_dict()的keys，里面全是base xxx，没有bottleneck

[cwhgn]

您好，这是由于bottleneck的layer是在后面创建的，把添加的这段代码，即

if model_path != '':
    print(self.state_dict().keys())
    param_dict = torch.load(model_path)
    for i in param_dict:
        if 'classifier' in i:
            continue
        self.state_dict()[i].copy_(param_dict[i])

放到下面代码后面即可。

SOLIDER-REID/model/make_model.py

Lines 222 to 226 in 8c08e1c

	self.bottleneck = nn.BatchNorm1d(self.in_planes)
	self.bottleneck.bias.requires_grad_(False)
	self.bottleneck.apply(weights_init_kaiming)
	self.dropout = nn.Dropout(self.dropout_rate)

[aliencaocao]

可以了，但是又有个新的问题：

Traceback (most recent call last):
  File "C:\Users\alien\Documents\PyCharm-Projects\TIL-2023\CV\SOLIDER-REID\train.py", line 84, in <module>
    do_train(
  File "C:\Users\alien\Documents\PyCharm-Projects\TIL-2023\CV\SOLIDER-REID\processor\processor.py", line 94, in do_train
    time_per_batch = (end_time - start_time) / (n_iter + 1)
UnboundLocalError: local variable 'n_iter' referenced before assignment

[cwhgn]

您看下是不是您读取数据的train_loader出了问题，导致没有进入下面这个for循环。

SOLIDER-REID/processor/processor.py

Line 51 in 8c08e1c

for n_iter, (img, vid, target_cam, target_view) in enumerate(train_loader):

[aliencaocao]

print(len(train_loader.dataset)) 显示是3，我现在就是在测试所以只放了三张图片：

[aliencaocao]

不过确实是没有进到loop里，我在排查为什么

[aliencaocao]

找到原因了，因为我一共就三个图片，但是batchsize是64，把batchsize降到3就可以了，谢谢大佬