You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello, I'm not sure if GCNet in mmdetection use ImageNet pretrained model (specifically GCNet + ResNet pretrained model) cause I saw in the GCNet/config, it only uses standard "torchvision://resnet50" in official mmedetection and "modelzoo://resnet50" in this repo as pretrained model, but in the original paper, the auther firstly trained on ImageNet and transfor to COCO detection task.
Would you explain does mmdetection use (ResNet pretrained model) or (ResNet+GC pretrained model)? if you only use ResNet pretrained model, why does COCO have such a large improvement?
The text was updated successfully, but these errors were encountered:
Hello, I'm not sure if GCNet in mmdetection use ImageNet pretrained model (specifically GCNet + ResNet pretrained model) cause I saw in the GCNet/config, it only uses standard "torchvision://resnet50" in official mmedetection and "modelzoo://resnet50" in this repo as pretrained model, but in the original paper, the auther firstly trained on ImageNet and transfor to COCO detection task.
Would you explain does mmdetection use (ResNet pretrained model) or (ResNet+GC pretrained model)? if you only use ResNet pretrained model, why does COCO have such a large improvement?
The text was updated successfully, but these errors were encountered: