Toytiny/CMFlow

But when I rerun the code, I get a RuntimeError:CUDA out of memory. I changed the size of the batch size in the RAFT/ train.py file, but it doesn't fix the problem

Toytiny opened this issue · 2 comments

Toytiny commented 
          But when I rerun the code, I get a RuntimeError:CUDA out of memory. I changed the size of the batch size in the RAFT/ train.py file, but it doesn't fix the problem

Originally posted by @syh1620 in #1 (comment)

Toytiny commented

If you want to reduce the memory consumption, please change the parameter RESIZE_SCALE TO 1/2 in https://github.com/Toytiny/CMFlow/blob/master/preprocess/utils/global_param.py. It will downsample the input image by 1/2 and upsample the optical flow result back after inference.