Is the convolution layer correct in ResUnetv4?

Hi, I have reproduced the performance of resUNet, but I have a question when looking at the model structure.

On line 114 of resnet.py, I think you need to change conv1d to conv2d.
It doesn't work with Python >= 2.0, and looking at the internal parameters, it doesn't look like this is intended.

However, this code works in pytorch 1.5.0, so if you have a reason to use conv1, please let me know.

Thank you.