Closed lyangfan closed 10 months ago
The reason: our codebase uses convnext's official definition (downstream_imagenet/models/convnext_official.py) rather than timm.ConvNeXt
.
Solution: import
this convnext_official.py file before timm.create_model('convnext_small')
. Our definitions and registrations in convnext_official.py will override and replace those of timm
.
Thanks for your reply!
I trained a convnext_small model using my own dataset. When I try to load the weights:
I got these output:
It seems that the keys are not aligned My timm version is 0.5.4
Looking forward to your reply Liu Yangfan 2023.9.28