目标检测实验问题记录（5）AttributeError: ‘RoIAlignRotated‘ object has no attribute ‘out_size

问题描述

训练Pointobb网络，模型使用了mmcv和mmdet框架

Traceback (most recent call last):
  File "tools/train_dist.py", line 192, in <module>
    main()
  File "tools/train_dist.py", line 181, in main
    train_detector(
  File "/data/lxy1/PointOBB/mmdet/apis/train.py", line 172, in train_detector
    runner.run(data_loaders, cfg.workflow)
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 136, in run
    epoch_runner(data_loaders[i], **kwargs)
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 53, in train
    self.run_iter(data_batch, train_mode=True, **kwargs)
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 31, in run_iter
    outputs = self.model.train_step(data_batch, self.optimizer,
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/parallel/distributed.py", line 63, in train_step
    output = self.module.train_step(*inputs[0], **kwargs[0])
  File "/data/lxy1/PointOBB/mmdet/models/detectors/base.py", line 241, in train_step
    losses = self(**data)
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
    return forward_call(*input, **kwargs)
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 119, in new_func
    return old_func(*args, **kwargs)
  File "/data/lxy1/PointOBB/mmdet/models/detectors/base.py", line 175, in forward
    return self.forward_train(img, img_metas, **kwargs)
  File "/data/lxy1/PointOBB/mmdet/models/detectors/PointOBB.py", line 636, in forward_train
    roi_losses_0, bbox_results, pseudo_boxes, dynamic_weight = self.roi_head.forward_train(self.stage, feat, img_metas_all,
  File "/data/lxy1/PointOBB/mmdet/models/roi_heads/PointOBB_head.py", line 719, in forward_train
    bbox_results = self._bbox_forward_train(x, proposal_list_base, proposals_list, proposals_valid_list,
  File "/data/lxy1/PointOBB/mmdet/models/roi_heads/PointOBB_head.py", line 738, in _bbox_forward_train
    bbox_results = self._bbox_forward(x, rois, gt_true_bboxes, stage)
  File "/data/lxy1/PointOBB/mmdet/models/roi_heads/PointOBB_head.py", line 795, in _bbox_forward
    bbox_feats = self.bbox_roi_extractor(
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
    return forward_call(*input, **kwargs)
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/runner/fp16_utils.py", line 208, in new_func
    return old_func(*args, **kwargs)
  File "/data/lxy1/PointOBB/mmdet/models/roi_heads/roi_extractors/rotate_single_level_roi_extractor.py", line 108, in forward
    out_size = self.roi_layers[0].out_size  # MMCV版本过低,故使用旧版本写法
  File "/data/lxy1/Anaconda3/envs/openmmlab/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in __getattr__
    raise AttributeError("'{}' object has no attribute '{}'".format(
AttributeError: 'RoIAlignRotated' object has no attribute 'out_size'

原因分析

报错的具体部分在于 PointOBB_head.py 文件中的 _bbox_forward 方法调用了 self.bbox_roi_extractor 的 forward 方法，并试图访问 RoIAlignRotated 对象的 out_size 属性，而这个属性不存在。

File "/data/lxy1/PointOBB/mmdet/models/roi_heads/roi_extractors/rotate_single_level_roi_extractor.py", line 108, in forward
    out_size = self.roi_layers[0].out_size  # MMCV版本过低,故使用旧版本写法

解决方案

尝试gpt给的建议

1、检查和更新依赖项： 确保 mmcv 和 mmdet 库是兼容的版本。

我的库版本：mmcv-full                 1.7.2

conda list中没有mmdet，在项目根目录有一个mmdet文件夹

所以这个步骤先跳过

2、确认属性是否存在

检查 RoIAlignRotated 对象是否有 out_size 属性，如果没有，需要调整代码以适应当前版本的实现。

检查方法：打印所有属性名

print(dir(self.roi_layers[0]))

结果显示，没有  out_size 属性，但是有output_size。

3、参考官方文档或示例代码：

查阅 mmcv 或 mmdet 官方文档，找到最新版本中 RoIAlignRotated 的正确使用方法。

这一步偷个懒，就不查了，应该就是用output_size替代 out_size。而且gpt给的修改示例也是output_size。

临时解决方法

out_size = self.roi_layers[0].output_size  # 使用新的属性名

成功解决