segmentation tasks

[youngbaldy]

when i want to use one_peace_vision.pt as pretrained model for vision tasks, i met this error:

• mmseg - INFO - load checkpoint from local path: /home/ONE-PEACE-main/one_peace_vision.pt Traceback (most recent call last): File "one_peace_vision/seg/train.py", line 243, in main() File "one_peace_vision/seg/train.py", line 203, in main model.init_weights() File "/data/OEM/lib/python3.8/site-packages/mmcv/runner/base_module.py", line 116, in init_weights m.init_weights() File "/home/ONE-PEACE-main/one_peace_vision/seg/mmseg_custom/models/backbones/onepeace.py", line 570, in init_weights state_dict = self.resize_abs_pos_embed(model) File "/home/ONE-PEACE-main/one_peace_vision/seg/mmseg_custom/models/backbones/onepeace.py", line 467, in resize_abs_pos_embed pos_embed_checkpoint = checkpoint['image_adapter.pos_embed'] KeyError: 'image_adapter.pos_embed'

i dont know how to handle this problem...

[simonJJJ]

Hi, follow link to extract vision branch model. For example:

python convert_to_vision.py /path/to/one-peace.pt /save_path/one-peace-vision.pkl

Then use the one-peace-vision.pkl for semantic segmentation.

Note that the one_peace_vision.pth is different from one-peace-vision.pkl in model names.

[youngbaldy]

oooook, i will try as u suggested, hope you have a nice day, 3Q~