it's too slow that use GPU

iscyy / yoloair

🔥🔥🔥 专注于YOLOv5，YOLOv7、YOLOv8、YOLOv9改进模型，Support to improve backbone, neck, head, loss, IoU, NMS and other modules🚀

https://github.com/iscyy/yoloair

GNU General Public License v3.0

2.51k stars 430 forks source link

it's too slow that use GPU #96

Closed Alioth5670 closed 2 years ago

Alioth5670 commented 2 years ago

Search before asking

[X] I have searched the YOLOAir issues and found no similar questions.

Question

When I train YOLOv5 model with yoloair, it's slower than the model in official yolov5, and Volatile GPU-Util less than 10% in my windows PC, with NVIDIA GeForce 2080Ti.

Additional

use command: python train.py --weights ' ' --cfg configs/yolov5-standard/yolov5s.yaml --img 640 --bitch-size 16

yang-0201 commented 2 years ago

maybe you shoud set "--wokers" to 8

Alioth5670 commented 2 years ago

parser.add_argument('--workers', type=int, default=8, help='max dataloader workers (per RANK in DDP mode)') In train.py, it is default that workers are set to 8. So does it have another way to deal with this case?

yang-0201 commented 2 years ago

try to set "--device" to 0

Alioth5670 commented 2 years ago

I have done the same thing before, but this method didn't work.

iscyy commented 2 years ago

The training of yolov5 is consistent with the official one, and the situation you said will not occur. Can you post your train.py parameter part?

Alioth5670 commented 2 years ago

OMG！ I confused the 'train.py' of official YOLOv5 and YOLOAIR, which were opened in the same windows. In the official version, 'workers' was set to 8, but not in yoloair. I'm sorry to trouble you. Thank you for your patience.