search

WongKinYiu / yolov7

Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
GNU General Public License v3.0
13.01k stars 4.12k forks source link

Question about YOLOR and YOLOv7 reported FPS #282

Open zeerandhawa opened 1 year ago

zeerandhawa commented 1 year ago

I am currently comparing inference speed of YOLOR-p6 and YOLOv7-w6 on GTX 1080 Ti.

Based on the reported results YOLOv7-w6 should have a higher FPS than YOLOR-p6.

However, using a batch size 2, YOLOv7-w6 is slower than YOLOR-p6. (YOLOv7-w6 has lower GPU utilization though).

Could you suggest why this is the case? Were the results in both repos reported on different GPUs?

Thank you for your help in advance.

AlexeyAB commented 1 year ago

Both were tested on the same nVidia GPU Tesla V100 using this Docker image https://github.com/WongKinYiu/yolov7#installation and this command:

python test.py --data data/coco.yaml --img 1280 --batch 1 --conf 0.001 --iou 0.65 --device 0 --weights yolov7-w6.pt --name yolov7_w6_1280_val
zeerandhawa commented 1 year ago

Thank you for your response.

Just to confirm, the reported FPS is only the model throughput time, right? It does not includes NMS time?

WongKinYiu commented 1 year ago

Yes, only model inference.

charming2992 commented 1 year ago

Can you tell me how to get the FPS? Run test.py to get the time for per image , then calculate it?