Convert Pytorch Batch Normalization weights to Tensorflow Batch Normalization

I have been working on converting ResNeXt50 weights from Pytorch into Tensorflow. I seem to have transferred all other weights from the convolution layers and fully connected head, however, I can't seem to correctly transfer the values inside the batch normalization layers. From what I understand, Pytorch batch normalization layers have a weight, bias, running_mean, and running_var. These correspond to the gamma, beta, moving_mean, and moving_variance in Tensorflow's batch normalization. I have inspected the weights of the batch normalization layer when I convert them into Tensorflow, and it seems that everything worked. However, when I go to make predictions, the output values of the Tensorflow batch normalization are way off from the output values of the Pytorch batch normalization. Is this discrepancy due to an incorrect conversion, or a difference in the behavior of batch normalization at inference in both libraries?