`KeyError: 'conv1_w'` When converting from Detectron2 to MMDetection

[justlike-prog]

Hi I would like to convert a Detectron2 into a MMDetection model. Running the detectron2pytorch.py script leads to an error though,

python detectron2pytorch.py model_final_f10217.pkl test123.pth 50 Traceback (most recent call last): File "detectron2pytorch.py", line 83, in <module> main() File "detectron2pytorch.py", line 79, in main convert(args.src, args.dst, args.depth) File "detectron2pytorch.py", line 47, in convert convert_conv_fc(blobs, state_dict, 'conv1', 'conv1', converted_names) File "detectron2pytorch.py", line 26, in convert_conv_fc state_dict[torch_name + '.weight'] = torch.from_numpy(blobs[caffe_name + KeyError: 'conv1_w'

[BIGWangYuDong]

what checkpoints did you use?

[kowshikthopalli]

Hi @BIGWangYuDong is detectron2pytorch.py for converting models trained in original Detectron or Detectron2? I have trained few object detection models in detectron2. I have recently seen that you include robustness checking as a part of your toolbox. Is there a possibility for me to directly evaluate my detectron2 models for benchmarking for robustness either by converting it to mmdet or other means?