QAT training backbone problem？

[betterhalfwzm]

when i train yolo v3 with darknet, qat training accuracy is ok, but replace the backbone network like vgg，and floating point training is ok,but qat training accuracy is very low @mathmanu (backbone): VGG( (stage0): VGGBlock( (conv): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1)) (bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (stage1): Sequential( (0): VGGBlock( (conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1)) (bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (1): VGGBlock( (conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) ) (stage2): Sequential( (0): VGGBlock( (conv): Conv2d(32, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1)) (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (1): VGGBlock( (conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (2): VGGBlock( (conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (3): VGGBlock( (conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) ) (stage3): Sequential( (0): VGGBlock( (conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1)) (bn): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (1): VGGBlock( (conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (2): VGGBlock( (conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (3): VGGBlock( (conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (4): VGGBlock( (conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (5): VGGBlock( (conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (6): VGGBlock( (conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) (7): VGGBlock( (conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) ) (stage4): Sequential( (0): VGGBlock( (conv): Conv2d(96, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1)) (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (activate): ReLU(inplace=True) ) ) )

@mathmanu