Solving ''CUDA out of memory'' -- Run FreeReg on RTX3060 (12G)

[Jackiezhou233]

RuntimeError: CUDA out of memory. Tried to allocate 968.00 MiB (GPU 0; 11.76 GiB total capacity; 8.87 GiB already allocated; 484.12 MiB free; 9.95 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

My device: RTX 3060 with 12GB capacity.

Is there someway to run this? Thanks!

[Jackiezhou233]

Solved by release zoe's gpu after preprocess, now it works by RTX 3060 with 12GB. Modify(demo.py):

import gc

    def run(self):
        # load feature
        self.load_data()

        # densify and zoe
        self.preprocess()
        del self.zoe
        gc.collect()  # trigger trash recycle (to del zoe)
        torch.cuda.empty_cache()  # release zoe's gpu

        # extract features
        self.extraction()
        del self.extractor
        gc.collect()
        torch.cuda.empty_cache()

        # registration
        torch.cuda.empty_cache()
        self.match()
        # evaluation
        self.eval()

[HpWang-whu]

Hi @Jackiezhou233 ,

Well done!

Yours,