Open shani-sony opened 1 year ago
@shani-sony That's understandable. Since the EfficientNet implementation we have is not based on torchvision, you can directly register torchvision's EfficientNet into MMDetection. This way, you can utilize their pre-trained weights.
I'm currently using this config to run detection models with EfficientNet as a backbone with the pre-trained model given in the config. When I try to use torchvision's
torchvision://efficientnet_bi
pre-trained backbones I get many mismatches (the params names appears to be different). Is there any way or config that support using these pre-trained models? Thank you :)