-
Notifications
You must be signed in to change notification settings - Fork 168
Open
Description
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?
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels