Pytorch to onnx to trt, BN layer output wrong result

[wangxuezhan10]

part of the model is as following.:

        nn.Conv2d(3, 64, 3, stride=1, padding=1),
        nn.BatchNorm2d(64),
        nn.ReLU(inplace=True),
        nn.MaxPool2d(3, 2, padding=1),
    when running the BN layer, output by pytorch and onnx is different.the

ckpt.t7 (43.9 MB)

[AastaLLL]

Hi,

Just want to confirm first.
Do you find the difference between PyTorch vs. ONNX or PyTorch vs. TensorRT?

If the difference occurs when converting the PyTorch model into ONNX.
It’s recommended to check this issue with the PyTorch team for better help.

Thanks.