Your Tiny yolo v3 implementation seems to be wrong

mystic123 / tensorflow-yolo-v3

Implementation of YOLO v3 object detector in Tensorflow (TF-Slim)

https://medium.com/@pawekapica_31302/implementing-yolo-v3-in-tensorflow-tf-slim-c3c55ff59dbe

Apache License 2.0

890 stars 353 forks source link

Your Tiny yolo v3 implementation seems to be wrong #50

Open LucasMahieu opened 5 years ago

LucasMahieu commented 5 years ago

Hy @mystic123 ,

It seems that you implementation is not exactly the same as the original one (which is leading to bad performance in detection).

Here are the clues : According to the original implementation here

You should uncomment these line in your code : here and here

Can you explain us why you have commented these lines ?

mystic123 commented 5 years ago

Hi, Thank's for your comment. Yolo v3 tiny implementations comes from PR and to be honest, I did not check it thoroughly. If you are 100% that it's wrong, please submit PR with a fix.

shenyingying commented 5 years ago

@LucasMahieu @mystic123 thanks for you post,but the .pb file can read,can't detection in office tensorflow andriod demo

nikolaevra commented 5 years ago

@LucasMahieu I looked into your concern there, but there is no error. The linear activation layer is implemeted inside of the _detection_layer() function.