Junyu Chen
Junyu Chen
Hi @tuffr5 , I tried VoxelMorph again using the same scripts that I provided for 11 epochs, and it seemed everything was working just fine:  Can you provide me...
@tuffr5 Yep. That definitely doesn't look right. I'm wondering whether this is due to some type of pytorch update that affects the spatial transformation function. However, it seemed like the...
@tuffr5 Hmm... Interesting. I'll try replicating your problem with 1.9 and 1.11 once I have more time. Thanks for letting me know.
Hi @lemoshu , I have translated @adalca 's MI code (https://github.com/adalca/neurite/blob/dev/neurite/tf/metrics.py) into Pytorch: https://github.com/junyuchen245/ViT-V-Net_for_3D_Image_Registration_Pytorch/blob/b6a9c884ecebb0b0126be7c1aae607050ec54d70/ViT-V-Net/losses.py#L288 I did some quick tests, and the code seems to be working. I hope this helps.
> @junyuchen245 This is cool -- would you mind also adding this to dev/neurite/torch/metrics.py as an open source PR? > > We could do it but it's helpful if the...
Sure thing. I'm glad that I could help.
Hi @timroelofs123 , I had a similar issue, but I double-checked the code, and it matched @adalca 's TensorFlow version. Anyways, I have been using this simplified version, and it's...
@timroelofs123 I'm glad to hear it works. I just uploaded the PyTorch version of the local MI loss (for non-overlapping local patches), which I implemented based on a previous TF...
Hi @Dududu233 , Sorry for the late reply. You can use [torch.load](https://pytorch.org/docs/stable/generated/torch.load.html) to load the pretrained weights (i.e., the entire .tar file) into the model. We additionally provided pre-trained weights...
Hi @tzayuan , It is found that the VoxelMorph script currently only compatible with PyTorch 1.7 [see here](https://github.com/junyuchen245/TransMorph_Transformer_for_Medical_Image_Registration/issues/15#issuecomment-1088012308). While I continue to investigate, you can test it on PyTorch 1.7...