TypeError: init_weights() missing 1 required positional argument: 'pretrained'

[apanand14]

File "tools/train.py", line 189, in
main()
File "tools/train.py", line 163, in main
model.init_weights()
File "C:\Users\topseven\anaconda3\envs\mmcv\lib\site-packages\mmcv\runner\base_module.py", line 117, in init_weights
m.init_weights()
TypeError: init_weights() missing 1 required positional argument: 'pretrained'

[zhouzaida]

Did you override the init_weights() method in your module?

[apanand14]

I think so but not sure. I think there is another model.init_weights() in train.py But When I comment it out then I'm able to train the model. I do not know is it right way or Am I missing something?

[zhouzaida]

What codebase are you using?

[apanand14]

mmdetection repository and training file from tools/train.py

[zhouzaida]

please provide your mmdet version and what config you are using?

[apanand14]

mmcv-full : 1.3.12
mmdet: 2.16.0

When I run with default mask_rcnn and config: mask_rcnn_r50_caffe_fpn_mstrain-poly_1x_coco.py, it runs fine.
Thank you for replying back

[zhouzaida]

so why the error happened? Did you modify the mask_rcnn model?

[apanand14]

No. Basically I'm using this custom implementation.
https://github.com/zhanggang001/RefineMask
If I run this then I probably got an error. If I comment model.init_weights(), then I'm able to run but I get some errors in test.py during evaluation. I talked to author about it and he said that better us the exact versions of mmcv-full and mmdet to reproduce the output. If you try with the same mmcv-full and mmdet versions and run test.py withput error then let me know please. Thank you in advance

[zhouzaida]

@hhaAndroid have a look

[hhaAndroid]

@apanand14 Please refer to https://github.com/open-mmlab/mmdetection/blob/master/docs/changelog.md#backwards-incompatible-changes. In v2.12.0 MMDetection, we have adopted init_cfg uniformly, you need to refer to open-mmlab/mmdetection#4750 for custom code. If you still can’t solve it, we will further help you solve it.

[apanand14]

Hello @hhaAndroid , I tried to solve this issue for custom code. But I'm getting same error no matter what I chnage. Is it possible that you look into and help me to solve it please? Here is the custom implementation I use on mmcv-full 1.3.12 and mmdet 2.16.0 . As you mentioned backward incompatibale changes causes above mentioned error. Can you help me to run this custom implementation on mmcv-full and mmdet (latest versions)?

https://github.com/zhanggang001/RefineMask
In this custom implementation there are two custom implementation are done.

1. refinemask/mmdet/models/roi_heads/refine_roi_head.py and
2. refinemask/mmdet/models/roi_heads/mask_heads/refine_mask_head.py
3. Config file: configs/refinemask/coco/r50-refinemask-1x.py

As I told you before, I use init_cfg in cfg file. Also, I do not get any error while I train with default Mask Rcnn. (Training and inference works well). In custom implementation, training works if I make a comment of model.init_weights() but it throws an error (mentioned in title) with it. Please help me out. Thank you in advance!!

[hhaAndroid]

If you do not want to modify the current code, I suggest that your mmdet version should not be higher than v2.12.0, because there is BC-Breaking.

[apanand14]

Thank you for the reply. I will try to downgrade mmdet version and check it out.

[zhouzaida]

hi @apanand14 , is there any progress?

[apanand14]

Hello @zhouzaida , I'm able to run the customized mmdetection by installing the lower versions. Thank you for your help. I will let you know once I change the code on the latest versions of the mmdetection and then it works or not.

[zhouzaida]

Hello @zhouzaida , I'm able to run the customized mmdetection by installing the lower versions. Thank you for your help. I will let you know once I change the code on the latest versions of the mmdetection and then it works or not.

hi @apanand14 , is there any progress?

[apanand14]

No, I'm focused on the training and optimizing stuffs with lower versions only. I will let you know in maximum weeks.