search

qqwweee / keras-yolo3

A Keras implementation of YOLOv3 (Tensorflow backend)
MIT License
7.14k stars 3.44k forks source link

about tiny yolov3 #179

Open sherlockchou86 opened 6 years ago

sherlockchou86 commented 6 years ago

hi, I used yolov3 weight (252 layers) to train my dataset and it works very well :) (about 4w images in my dataset).

now i try to use tiny yolov3 weight to train my dataset because i need improve the process speed when do the predicting(about 0.036s per image on GTX 1080 GPU so far), but i have used tiny yolo model(converted using convert.py script), found the performace is not good, some targets cannot be detected. will the performace increase after my training?

so i want to speed up my predicting speed and keep a good performace , anybody give me advice?thanks!

below is my screenshot with full yolov3 weights(252 layers)

sherlockchou86 commented 6 years ago

I think the performace for tiny yolov3 is not as good as official website said :(

WDF1994 commented 5 years ago

Thank you. Can you share your code?e-mile：962396836@qq.com Thanks

bujianyiwang commented 4 years ago

@sherlockchou86 hello，can you share the code how you put id on each object, Thanks!