search

WongKinYiu / ScaledYOLOv4

Scaled-YOLOv4: Scaling Cross Stage Partial Network
GNU General Public License v3.0
2.02k stars 575 forks source link

Speed comparasion with yolov5 s/m/l? #69

Open lucasjinreal opened 3 years ago

lucasjinreal commented 3 years ago

Speed comparasion with yolov5 s/m/l?

WongKinYiu commented 3 years ago

there are no reported batch 1 latency for those models. our batch 1 fps are listed in the table. value with * means nms is included, while without means only model inference. image

dsantiago commented 3 years ago

@WongKinYiu is it on RTX2080ti like in the Darknet README?

WongKinYiu commented 3 years ago

@dsantiago in paper, v100. by the way, efficientdet-d0 should be 97*.

WongKinYiu commented 3 years ago

@jinfagang could you reproduce the speed of yolov5 models?

i download the models and inference on v100 gpu with command: Reproduce speed by python test.py --data coco.yaml --img 640 --conf 0.25 --iou 0.45

it gets: yolov5s: 1.2/0.8/2.0 ms yolov5m: 2.4/0.8/3.2 ms yolov5l: 3.9/0.7/4.6 ms yolov5x: 6.8/0.9/7.6 ms

which are far slower than the value listed on the table. and slower than yolov4 models when running same command on same machine.