CPU memory leak + cuDNN error when using the GPU with preprocess None
connorlee77 opened this issue · comments
Two issues I've been running into when attempting to perform inference with the test.py script:
- When running with GPU, with no preprocessing (i.e. input shape is size of input image), I run into a CUDNN_STATUS_INTERNAL_ERROR. Note that the code runs fine on the CPU except for the below:
- When running on CPU, the memory leaks tremendously to the point where the process gets killed on a machine with 128GB ram.
Any ideas how to fix either issues?
I am having the same issue with CUDNN_STATUS_INTERNAL_ERROR. Did you manage to find a solution?
No, I was not able to. That said, I didn't spend that much time trying to debug this. Still interested in why the memory leak is occuring though as to avoid it in future code.
btw, i did this stuff again and worked around this by just letting it do its default resizing to a square crop and then resizing back to the original dimensions after inference.