Hi,
Thank you for your great work on this project. I noticed an issue with loading the pre-trained checkpoint in the code. Currently, the load_state_dict method is being used to load the checkpoint, but it doesn't handle the key 'model' in the gf_detector_l6o256.pth file. Therefore, I suggest modifying the code to use the following:
Additionally, the pre-trained weights have the prefix module.backbone_net, so we need to remove it to load the weights properly.
Regarding the strict=False argument, it is currently being used in the code, which means the weights are being randomly initialized instead of using the pre-trained checkpoint. Is that intended?
Hi, Thank you for your great work on this project. I noticed an issue with loading the pre-trained checkpoint in the code. Currently, the load_state_dict method is being used to load the checkpoint, but it doesn't handle the key 'model' in the gf_detector_l6o256.pth file. Therefore, I suggest modifying the code to use the following:
Change:
to:
Additionally, the pre-trained weights have the prefix module.backbone_net, so we need to remove it to load the weights properly.
Regarding the strict=False argument, it is currently being used in the code, which means the weights are being randomly initialized instead of using the pre-trained checkpoint. Is that intended?
Best regards.