John1231983

Hi. Because i am using mura 1.1 so some files are missing comparison with mura 1.0. I guess problem is that `image = pil_loader(study_path + 'image%s.png' % (i+1)) ` ....

I remember i download it in kaggle in the second author website. But i did not remember exactly.

Hi you should smooth on deformation field , not warp image.

And i think the smooth loss is wrong implementation

No. y is warped/moved image. y true is fixed image. I think this project is wrong to reproduce result. You can check tf version. Smooth must be in displacement vector

@Hsankesara : This is unit test for smooth loss in 3D. Hope it help https://colab.research.google.com/drive/1GJl1zWxTPF4KyHwiqK8elraz0xqb855R For spatial transformation layer, could we simple use grid_sample() function in pytorch, instead of write...

Ok. Now both loss are close to tensorflow version. You csn use it

I think the first conv should be conv2d. Am I right? The correct version likes ``` self.spatial_conv = nn.Conv2d(in_channels, intermed_channels, kernel=3, stride=1, padding=1, bias=bias) self.bn = nn.BatchNorm2d(intermed_channels) self.relu = nn.ReLU()...

+1

@lixiangMindSpore do you find any way to convert official FGD to the yolox code?