fengjianhui158
commented
2 months ago 是我在resnet.py 这个文件中 改代码的问题，
if dcn_config is not None: //这行代码是新增加的
with_dcn = dcn_config.get("stage_with_dcn", False)
if with_dcn:
deformable_groups = dcn_config.get("deformable_groups", 1)
with_modulated_dcn = dcn_config.get("with_modulated_dcn", False)
self.conv2 = DFConv2d(
bottleneck_channels,
bottleneck_channels,
with_modulated_dcn=with_modulated_dcn,
kernel_size=3,
stride=stride_3x3,
groups=num_groups,
dilation=dilation,
deformable_groups=deformable_groups,
bias=False
)
else:
self.conv2 = Conv2d(
bottleneck_channels,
bottleneck_channels,
kernel_size=3,
stride=stride_3x3,
padding=dilation,
bias=False,
groups=num_groups,
dilation=dilation
)
nn.init.kaiming_uniform_(self.conv2.weight, a=1)但是，如果不加这个判断，则会报 如下错误
Traceback (most recent call last):
File "tools/train_net.py", line 255, in
我使用 docker pull pengchuanzhang/pytorch:ubuntu20.04_torch1.9-cuda11.3-nccl2.9.9 这镜像文件构建了训练环境； 然后进行 COCO Fine-Tuning 微调；
_coco.yaml 的配置信息如下： MODEL: META_ARCHITECTURE: "GeneralizedRCNN" WEIGHT: "swin_tiny_patch4_window7_224.pth" BACKBONE: FREEZE_CONV_BODY_AT: -1
use for grounding model
TEST: DURING_TRAINING: False IMS_PER_BATCH: 4 EVAL_TASK: "detection"
DATASETS: TRAIN: ("coco_2017_train", "coco_2017_train") TEST: ("coco_2017_test", ) DISABLE_SHUFFLE: true SOLVER: BASE_LR: 0.0001 LANG_LR: 0.00001 STEPS: (0.67, 0.89) MAX_EPOCH: 24 IMS_PER_BATCH: 4 USE_AMP: True FIND_UNUSED_PARAMETERS: False 然后执行 python tools/train_net.py --config-file "configs/pretrain/_coco.yaml" --skip-test 但是却报错，如下： Traceback (most recent call last): File "tools/train_net.py", line 255, in
main()
File "tools/train_net.py", line 248, in main
model = train(cfg=cfg,
File "tools/train_net.py", line 129, in train
do_train(
File "/workspace/glip/maskrcnn_benchmark/engine/trainer.py", line 123, in do_train
loss_dict = model(images, targets)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, kwargs)
File "/workspace/glip/maskrcnn_benchmark/modeling/detector/generalized_rcnn.py", line 107, in forward
x, result, detector_losses = self.roi_heads(features, proposals, targets)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, *kwargs)
File "/workspace/glip/maskrcnn_benchmark/modeling/roi_heads/init.py", line 28, in forward
x, detections, loss_box = self.box(features, proposals, targets)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(input, kwargs)
File "/opt/conda/lib/python3.8/site-packages/torch/cuda/amp/autocast_mode.py", line 233, in decorate_fwd
return fwd(*_cast(args, cast_inputs), _cast(kwargs, cast_inputs))
File "/workspace/glip/maskrcnn_benchmark/modeling/roi_heads/box_head/box_head.py", line 48, in forward
x = self.feature_extractor(features, proposals)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, *kwargs)
File "/workspace/glip/maskrcnn_benchmark/modeling/roi_heads/box_head/roi_box_feature_extractors.py", line 96, in forward
x = self.head(x)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(input, kwargs)
File "/workspace/glip/maskrcnn_benchmark/modeling/backbone/resnet.py", line 228, in forward
x = getattr(self, stage)(x)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, *kwargs)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/container.py", line 139, in forward
input = module(input)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(input, **kwargs)
File "/workspace/glip/maskrcnn_benchmark/modeling/backbone/resnet.py", line 391, in forward
out = self.conv2(out)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in getattr
raise AttributeError("'{}' object has no attribute '{}'".format(
AttributeError: 'BottleneckWithFixedBatchNorm' object has no attribute 'conv2'
希望能够指点解决此问题，谢谢！