search

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
893 stars 353 forks source link

the binding box is incorrect #78

Open wangnd opened 5 years ago

wangnd commented 5 years ago

result-yolov3-pb 20190522-194100-linux_14000-pb

the same process with convert_weights_pb.py . the first is from yolov3.weight, the second is from my yolov3 train weights. @mystic123

wangnd commented 5 years ago

my yolov3 train weights include person except dog and horse. when I use darknet.exe detector test to predict the same image , the binding box is right as the first image. It really makes me puzzling

wangnd commented 5 years ago

not only the person class, all other class has the same problem @mystic123 ,please check the convert code, I think maybe it has bug

lucy3589 commented 5 years ago

not only the person class, all other class has the same problem @mystic123 ,please check the convert code, I think maybe it has bug

Have you solve it? If you sloved it, could you tell me? Thanks very much!

lucy3589 commented 5 years ago

not only the person class, all other class has the same problem @mystic123 ,please check the convert code, I think maybe it has bug

Could you solved it? How do you do ?

computervisionlearner commented 5 years ago

not only the person class, all other class has the same problem @mystic123 ,please check the convert code, I think maybe it has bug

Could you solved it? How do you do ?

not only the person class, all other class has the same problem @mystic123 ,please check the convert code, I think maybe it has bug

I think you change the anchor size by kmeans. You shoud change yolo_v3.py too.