cuda out of memory

[qgh1223]

I met cuda out of memory when using 2080 with 11G memory.Is there any way to solve this problem?

[huanglianghua]

Hi, as a reference, we executed our experiments on 4 TitanX GPUs with 12G memory each, and the training costs around 8~10G memory per GPU.

If the out-of-memory error still occurs, you can use a smaller max_instances in configs/qg_rcnn_r50_fpn.py, e.g., from 8 to 6.

Alternatively, you can also add a --fp16 flag in your training script, which will however slightly reduce the performance by ~1%.