search

ghimiredhikura / Complex-YOLOv3

PyTorch implementation of Complex-YOLO paper with YoloV3
GNU General Public License v3.0
190 stars 69 forks source link

Inference speed #5

Open wsl1014 opened 5 years ago

wsl1014 commented 5 years ago

I want to know the speed of your detection... I ran the test_both_side_detection.py on the PowerEdge-R730, the fps is about 4... Looking forward to your reply, thanks!

ghost commented 5 years ago

@wsl1014 , can you pls tell me how @ghimiredhikura has used anchor sizes to get good accuracy. Kindly help

ghimiredhikura commented 5 years ago

Hi, @wsl1014,

Yes in the current source code 4 fps that you are getting is correct. Actually this is end to end speed, bev creation, run through yolove and post-processing. BEV generation function is implemented in cpu and most of the time is consumed at this part. If you can replace this function with GPU implementation you will get real time. Check out #1 issue.