change batch size

[danielzhangau]

After reviewing source code and modified segmentation parts, training raised the following error: RuntimeError: CUDA out of memory. Tried to allocate 24.00 MiB (GPU 0; 10.89 GiB total capacity; 9.69 GiB already allocated; 25.56 MiB free; 9.73 GiB reserved in total by PyTorch) I changed batch size from 24 to 8, solved this issue. But what will be the effect of doing this?

[JamesLeeeeeee]

modify in ./lib/config/default.py at _C.TEST.BATCH_SIZE_PER_GPU