After loss.backward()，grad not updated

[wlf121]

The gradient of these layers is None every time

module.recognition_network.visual_backbone_twostream.rgb_stream.backbone.base.0.conv_s.weight has None grad module.recognition_network.visual_backbone_twostream.rgb_stream.backbone.base.0.bn_s.weight has None grad module.recognition_network.visual_backbone_twostream.rgb_stream.backbone.base.0.bn_s.bias has None grad module.recognition_network.visual_backbone_twostream.rgb_stream.backbone.base.0.conv_t.weight has None grad module.recognition_network.visual_backbone_twostream.rgb_stream.backbone.base.0.bn_t.weight has None grad module.recognition_network.visual_backbone_twostream.rgb_stream.backbone.base.0.bn_t.bias has None grad

[wlf121]

This is the problem in both training.py and prediction.py