When the graphics card 1 is being used by others and the graphics card 2 is not being used, the video card 2 cannot be called for training, resulting in an overflow error

[lushbng]

When there are 4 g of video memory left on the first video card and 11 g of video memory left on the second video card, you can train with batch = 1; when the batch is set to 8, the video memory is not enough, but the video memory of the second video card is not used. The command is as follows python main.py tracking --exp_ id mot17_ half --dataset mot --dataset_ version 17halftrain --pre_ hm --ltrb_ amodal --same_ aug --hm_ disturb 0.05 --lost_ disturb 0.4 --fp_ disturb 0.1 --gpus 0,1 --load_ model ../models/ crowdhuman.pth --batch_ size 8

[lushbng]

CUDA out of memory. Tried to allocate 256.00 MiB (GPU 0; 10.76 GiB total capacity; 1.94 GiB already allocated; 218.75 MiB free; 2.23 MiB cached) The default is that the 0 GPU cannot be changed with the command

[lushbng]

The above error is obtained by using the following command

python main.py tracking --exp_id mot17_half --dataset mot --dataset_version 17halftrain --pre_hm --ltrb_amodal --same_aug --hm_disturb 0.05 --lost_disturb 0.4 --fp_disturb 0.1 --gpus 1 --load_model ../models/crowdhuman.pth