An open source object detection toolbox based on PyTorch

你好， 我下载的示例模型，采用默认配置参数，导出onnx模型

python tools/deployment/pytorch2onnx.py configs/picodet/picodet_s_416_coco.py configs/picodet/picodet_s_416_coco.pth   --output-file /data/workspace/mmdet_picodet_s_416_coco.onnx

导出 onnx模型时报错：

File "/home/zxx/code/picodet_repro/mmdet/models/dense_heads/base_dense_head.py", line 427, in onnx_export
    assert cls_score.size()[-2:] == bbox_pred.size()[-2:]

然后发现好像是picodet_head.py 里195 行的维度没有对齐。将代码修改如下

if torch.onnx.is_in_onnx_export():
            # cls_score = (
            #     torch.sigmoid(cls_score)
            #     .reshape(1, self.num_classes, -1)
            #     .permute(0, 2, 1)
            # )
            # bbox_pred = bbox_pred.reshape(1, (self.reg_max + 1) * 4, -1).permute(
            #     0, 2, 1
            # )

            bbox_pred = bbox_pred.permute(0, 2, 3, 1).reshape(1, -1, 4 * (self.reg_max + 1))
            pred_corners = self.integral(bbox_pred)
            bbox_pred = pred_corners.reshape(1,cls_feat.shape[-2],cls_feat.shape[-1], 4).permute(0,3,1,2)

return cls_score, bbox_pred

但依然会报错：

load checkpoint from local path: configs/picodet/picodet_s_416_coco.pth
The model and loaded state dict do not match exactly

unexpected key in source state_dict: bbox_head.p3_feat.scale_reg, bbox_head.p4_feat.scale_reg, bbox_head.p5_feat.scale_reg, bbox_head.p6_feat.scale_reg, bbox_head.p7_feat.scale_reg

missing keys in source state_dict: bbox_head.gfl_reg.0.weight, bbox_head.gfl_reg.0.bias, bbox_head.gfl_reg.1.weight, bbox_head.gfl_reg.1.bias, bbox_head.gfl_reg.2.weight, bbox_head.gfl_reg.2.bias, bbox_head.gfl_reg.3.weight, bbox_head.gfl_reg.3.bias
/home/zxx/code/picodet_repro/mmdet/models/dense_heads/base_dense_head.py:407: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
  cfg.get('nms_pre', -1), device=device, dtype=torch.long)

...

Traceback (most recent call last):
  File "tools/deployment/pytorch2onnx.py", line 345, in <module>
    skip_postprocess=args.skip_postprocess)
  File "tools/deployment/pytorch2onnx.py", line 100, in pytorch2onnx
    dynamic_axes=dynamic_axes)
  File "/home/zxx/miniconda3/envs/mmdet/lib/python3.7/site-packages/torch/onnx/__init__.py", line 276, in export
    custom_opsets, enable_onnx_checker, use_external_data_format)
  File "/home/zxx/miniconda3/envs/mmdet/lib/python3.7/site-packages/torch/onnx/utils.py", line 94, in export
    use_external_data_format=use_external_data_format)
  File "/home/zxx/miniconda3/envs/mmdet/lib/python3.7/site-packages/torch/onnx/utils.py", line 701, in _export
    dynamic_axes=dynamic_axes)
  File "/home/zxx/miniconda3/envs/mmdet/lib/python3.7/site-packages/torch/onnx/utils.py", line 503, in _model_to_graph
    _export_onnx_opset_version)
RuntimeError: number of dims don't match in permute

请问大佬有没有遇到这个问题，如何解决。

Traceback (most recent call last): File "tools/train.py", line 185, in main() File "tools/train.py", line 158, in main test_cfg=cfg.get('test_cfg')) File "/home/zijing.wang/.local/lib/python3.6/site-packages/mmdet/models/builder.py", line 59, in build_detector cfg, default_args=dict(train_cfg=train_cfg, test_cfg=test_cfg)) File "/home/zijing.wang/.local/lib/python3.6/site-packages/mmcv/utils/registry.py", line 213, in build return self.build_func(*args, **kwargs, registry=self) File "/home/zijing.wang/.local/lib/python3.6/site-packages/mmcv/cnn/builder.py", line 27, in build_model_from_cfg return build_from_cfg(cfg, registry, default_args) File "/home/zijing.wang/.local/lib/python3.6/site-packages/mmcv/utils/registry.py", line 46, in build_from_cfg f'{obj_type} is not in the {registry.name} registry') KeyError: 'PicoDet is not in the models registry'

Hello,

I am trying the code demo/image_demo.py . while loading the pretrained weights I encountered the following error

load checkpoint from local path: demo/picodet_s_320.26.9.pth
The model and loaded state dict do not match exactly

unexpected key in source state_dict: ema_backbone_conv1_conv_weight, ema_backbone_conv1_bn_weight, ema_backbone_conv1_bn_bias, ema_backbone_conv1_bn_running_mean, ema_backbone_conv1_bn_running_var, ema_backbone_conv1_bn_num_batches_tracked, ema_backbone_2_1_conv_dw_1_conv_weight, ema_backbone_2_1_conv_dw_1_bn_weight, ema_backbone_2_1_conv_dw_1_bn_bias, ema_backbone_2_1_conv_dw_1_bn_running_mean, ema_backbone_2_1_conv_dw_1_bn_running_var, ema_backbone_2_1_conv_dw_1_bn_num_batches_tracked, ema_backbone_2_1_conv_linear_1_conv_weight, ema_backbone_2_1_conv_linear_1_bn_weight, ema_backbone_2_1_conv_linear_1_bn_bias, ema_backbone_2_1_conv_linear_1_bn_running_mean, ema_backbone_2_1_conv_linear_1_bn_running_var, ema_backbone_2_1_conv_linear_1_bn_num_batches_tracked, ema_backbone_2_1_conv_pw_2_conv_weight, ema_backbone_2_1_conv_pw_2_bn_weight, ema_backbone_2_1_conv_pw_2_bn_bias, ema_backbone_2_1_conv_pw_2_bn_running_mean, ema_backbone_2_1_conv_pw_2_bn_running_var, ema_backbone_2_1_conv_pw_2_bn_num_batches_tracked, ema_backbone_2_1_conv_dw_2_conv_weight, ema_backbone_2_1_conv_dw_2_bn_weight, ema_backbone_2_1_conv_dw_2_bn_bias,
.....................................................

I tried the following config and checkpoint file

config_file = 'configs/picodet/picodet_s_320_coco.py'
checkpoint_file = 'picodet_s_320.26.9.pth'

Any suggestions how to fix this ? Thanks.

Thanks for your error report and we appreciate it a lot.

Checklist

1. I have searched related issues but cannot get the expected help.
2. I have read the FAQ documentation but cannot get the expected help.
3. The bug has not been fixed in the latest version.

Describe the bug A clear and concise description of what the bug is.

Reproduction

1. What command or script did you run?

A placeholder for the command.

2. Did you make any modifications on the code or config? Did you understand what you have modified?
3. What dataset did you use?

Environment

1. Please run python mmdet/utils/collect_env.py to collect necessary environment information and paste it here.
2. You may add addition that may be helpful for locating the problem, such as
   • How you installed PyTorch [e.g., pip, conda, source]
   • Other environment variables that may be related (such as $PATH, $LD_LIBRARY_PATH, $PYTHONPATH, etc.)

Error traceback If applicable, paste the error trackback here.

A placeholder for trackback.

Bug fix If you have already identified the reason, you can provide the information here. If you are willing to create a PR to fix it, please also leave a comment here and that would be much appreciated!

Describe the feature 我也在复现picodet的网络，但是loss在训练集快速降到很小，但是验证集上的loss却比较大，map很低，由于是从paddlepaddle中抽取的部分代码，可能在训练上存在问题，也可能代码还有细节没注意到，希望博主能分享下训练的具体过程。 顺便提一下，预训练模型能在百度云上也放一个不？感谢 Motivation A clear and concise description of the motivation of the feature. Ex1. It is inconvenient when [....]. Ex2. There is a recent paper [....], which is very helpful for [....].

Related resources If there is an official code release or third-party implementations, please also provide the information here, which would be very helpful.

Additional context Add any other context or screenshots about the feature request here. If you would like to implement the feature and create a PR, please leave a comment here and that would be much appreciated.