cogsys-tuebingen / mobilestereonet

Lightweight stereo matching network based on MobileNet blocks
Apache License 2.0
240 stars 47 forks source link

How I can reduce GPU memory usage #10

Closed Qjizhi closed 2 years ago

Qjizhi commented 3 years ago

my device is 3080 16GB, but with the original code, I can only train with batch_size=1, otherwise, cuda out of memory.

I tried with Mixed Precision, and I can increase batch_size to 2, how can I train with large batch size without dropping accuracy.

samuel-woerz commented 2 years ago

As mentioned in the mail, we used multiple GPUs to enable bigger batch sizes.