Why do the resnet50 results of arm neon cnn not correspond to pytorch?

[NNUCJ]

1. MY Goal:
      Using arm neon CNN to build resnet50 and compare the results with Pytorch. I have saved weights of Pytorch as numpy, and loaded the corresponding weights on Neon CNN.

2. Problems:
      The value of Feature Map cannot be aligned with Pytorch starting from the 3x3 conv2d() layer of the first bottleneck（The section between # comments in the network structure of the following code）.

ResNet(
  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (relu): ReLU(inplace=True)
  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
  (layer1): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      ##########################
      [error begin]
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    ############################
      (downsample): Sequential(
        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )

3. Test Script
     Using the same input data and convolutional weights(download link) to compare the output results of conv2d in acl and pytorch。

• pytorch

input = np.load("input.npy")
conv2_weights = np.load("layer1_0_conv2_weight.npy")
state_dict = {"weight": torch.from_numpy(conv2_weights)}
conv2d_model = nn.Conv2d(64, 64, kernel_size=3, stride=1,padding=1, groups=1, bias=False, dilation=1)
conv2d_model.load_state_dict(state_dict)
test_out = conv2d_model(torch.from_numpy(input).unsqueeze(0))

• arm neon cpp