Hi, thanks for your repo!
When I run python run.py (input is a video) on a 4 GPU machine, the code will take up 9872Mb per GPU. However, when I try to downsize the input by adjusting width=608, height=608 (to 160 for example) in darkflow/cfg/yolo.cfg, the memory usage doesn't change at all. I also try to comment the tracking code, the memory usage doesn't change. I think the memory usage is too large for detection.
Hi, thanks for your repo! When I run
python run.py
(input is a video) on a 4 GPU machine, the code will take up 9872Mb per GPU. However, when I try to downsize the input by adjustingwidth=608, height=608
(to 160 for example) indarkflow/cfg/yolo.cfg
, the memory usage doesn't change at all. I also try to comment the tracking code, the memory usage doesn't change. I think the memory usage is too large for detection.