search

cwmok / C2FViT

This is the official Pytorch implementation of "Affine Medical Image Registration with Coarse-to-Fine Vision Transformer" (CVPR 2022), written by Tony C. W. Mok and Albert C. S. Chung.
MIT License
132 stars 4 forks source link

How does the batchsize and lr influence the performance #6

Closed xyimaging closed 1 year ago

xyimaging commented 1 year ago

First of all, thanks so much for your good work!

I'm training on some 256x256x256 images and I find there is still a lot of free memory, so I'd like to use larger batchsize and lr to speed up the training. It would be great if you have any ablation study or insight on these 2 factors' influence. Thanks!

cwmok commented 1 year ago

Hi @xyimaging,

I didn't try using a larger batch size/lr. Empirically, I found that the current learning rate work well for many CNN-based registration networks.

Besides, the scale_factor of the input image also greatly affects the registration accuracy. Feel free to let me know if you have any interesting results. :D

xyimaging commented 1 year ago

Thanks a lot for your sharing!