Confused by the apply_fix(self) function. Would mind telling us why you set it?

[Ai-is-light]

Confused by the apply_fix(self) function. Would mind telling us why you set it? def apply_fix(self):

1. fix bn

    # 2. fix conv1 conv2
    for param in self.resnet.conv1.parameters():
        param.requires_grad = False
    for param in self.resnet.layer1.parameters():
        param.requires_grad = False

@Xiangyu-CAS Thanks

[Xiangyu-CAS]

In some past works, such as Faster-RCNN, R-FCN, they fixed the first blocks of backbones. They think the first two blocks are in charge of extracting low level features such as lines and edges, and it could be used in all kind of problems without fine-tuning.