R_50_FPN的module和各層維度

(backbone): Sequential(
    (body): ResNet(
      (stem): StemWithFixedBatchNorm(
        (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
        (bn1): FrozenBatchNorm2d()
      )
      (layer1): Sequential(
        (0): BottleneckWithFixedBatchNorm(
          (downsample): Sequential(
            (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
            (1): FrozenBatchNorm2d()
          )
          (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (1): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (2): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
      )
      (layer2): Sequential(
        (0): BottleneckWithFixedBatchNorm(
          (downsample): Sequential(
            (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
            (1): FrozenBatchNorm2d()
          )
          (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (1): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (2): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (3): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
      )
      (layer3): Sequential(
        (0): BottleneckWithFixedBatchNorm(
          (downsample): Sequential(
            (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
            (1): FrozenBatchNorm2d()
          )
          (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (1): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (2): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (3): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (4): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (5): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
      )
      (layer4): Sequential(
        (0): BottleneckWithFixedBatchNorm(
          (downsample): Sequential(
            (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
            (1): FrozenBatchNorm2d()
          )
          (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (1): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
        (2): BottleneckWithFixedBatchNorm(
          (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn1): FrozenBatchNorm2d()
          (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
          (bn2): FrozenBatchNorm2d()
          (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
          (bn3): FrozenBatchNorm2d()
        )
      )
    )
    (fpn): FPN(
      (fpn_inner1): Conv2d(256, 490, kernel_size=(1, 1), stride=(1, 1))
      (fpn_layer1): Conv2d(490, 10, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (fpn_inner2): Conv2d(512, 490, kernel_size=(1, 1), stride=(1, 1))
      (fpn_layer2): Conv2d(490, 10, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (fpn_inner3): GlobalContextModule(
        (col_max): Conv2d(1024, 256, kernel_size=(15, 1), stride=(1, 1), padding=(7, 0))
        (col): Conv2d(256, 490, kernel_size=(1, 15), stride=(1, 1), padding=(0, 7))
        (row_max): Conv2d(1024, 256, kernel_size=(1, 15), stride=(1, 1), padding=(0, 7))
        (row): Conv2d(256, 490, kernel_size=(15, 1), stride=(1, 1), padding=(7, 0))
      )
      (fpn_layer3): Conv2d(490, 10, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (fpn_inner4): GlobalContextModule(
        (col_max): Conv2d(2048, 256, kernel_size=(15, 1), stride=(1, 1), padding=(7, 0))
        (col): Conv2d(256, 490, kernel_size=(1, 15), stride=(1, 1), padding=(0, 7))
        (row_max): Conv2d(2048, 256, kernel_size=(1, 15), stride=(1, 1), padding=(0, 7))
        (row): Conv2d(256, 490, kernel_size=(15, 1), stride=(1, 1), padding=(7, 0))
      )
      (fpn_layer4): Conv2d(490, 10, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (top_blocks): LastLevelMaxPool()
    )
  )
  (rpn): RPNModule(
    (anchor_generator): AnchorGenerator(
      (cell_anchors): BufferList()
    )
    (head): RPNHead(
      (conv): Conv2d(10, 10, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (cls_logits): Conv2d(10, 3, kernel_size=(1, 1), stride=(1, 1))
      (bbox_pred): Conv2d(10, 12, kernel_size=(1, 1), stride=(1, 1))
    )
    (box_selector_train): RPNPostProcessor()
    (box_selector_test): RPNPostProcessor()
  )
  (roi_heads): CombinedROIHeads(
    (box): ROIBoxHead(
      (feature_extractor): FPN2MLPFeatureExtractor(
        (pooler): Pooler(
          (poolers): ModuleList(
            (0): ROIAlign(output_size=(7, 7), spatial_scale=0.25, sampling_ratio=2)
            (1): ROIAlign(output_size=(7, 7), spatial_scale=0.125, sampling_ratio=2)
            (2): ROIAlign(output_size=(7, 7), spatial_scale=0.0625, sampling_ratio=2)
            (3): ROIAlign(output_size=(7, 7), spatial_scale=0.03125, sampling_ratio=2)
          )
        )
        (fc6): Linear(in_features=490, out_features=1024, bias=True)
        (fc7): Linear(in_features=1024, out_features=1024, bias=True)
      )
      (predictor): FPNPredictor(
        (cls_score): Linear(in_features=1024, out_features=2, bias=True)
        (bbox_pred): Linear(in_features=1024, out_features=8, bias=True)
      )
      (post_processor): PostProcessor()
    )
    (keypoint): ROIKeypointHead(
      (feature_extractor): KeypointRCNNFeatureExtractor(
        (pooler): Pooler(
          (poolers): ModuleList(
            (0): ROIAlign(output_size=(14, 14), spatial_scale=0.25, sampling_ratio=2)
            (1): ROIAlign(output_size=(14, 14), spatial_scale=0.125, sampling_ratio=2)
            (2): ROIAlign(output_size=(14, 14), spatial_scale=0.0625, sampling_ratio=2)
            (3): ROIAlign(output_size=(14, 14), spatial_scale=0.03125, sampling_ratio=2)
          )
        )
        (conv_fcn1): Conv2d(10, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
        (conv_fcn2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
        (conv_fcn3): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
        (conv_fcn4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
        (conv_fcn5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
        (conv_fcn6): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
        (conv_fcn7): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
        (conv_fcn8): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      )
      (predictor): KeypointRCNNPredictor(
        (kps_score_lowres): ConvTranspose2d(256, 17, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
      )
      (post_processor): KeypointPostProcessor()
    )
  )
)