Cuda Out of Memory when doing Inference
marcobarbierato opened this issue · comments
I wanted to try this model to inference single images without validation.
So I tried modifying the option.yml files appropriately following the implementation in HAT (link).
However, even with small images of 300x300, 40GBs of GPU RAM are not enough for the model.
So I tried implementing tile-based mode following the implementation in HAT link, but even with smaller tile sizes I still run out of memory.
How much memory does the model need? Am I doing something wrong? Thanks.