search

xiaochus / YOLOv3

Keras implementation of yolo v3 object detection.
MIT License
603 stars 263 forks source link

Some issues about box decode #1

Closed xiaohu2015 closed 6 years ago

xiaohu2015 commented 6 years ago

I found that you use: col = np.tile(np.arange(1, grid_w + 1), grid_w).reshape(-1, grid_w) row = np.tile(np.arange(1, grid_h + 1).reshape(-1, 1), grid_h)

It means you use offsets of [1, W] and [1, H], But why you not use [0, W-1] and [0, H-1]?

xiaochus commented 6 years ago

@xiaohu2015 It should be [0, w-1] according to the implementation of the source code. But when i using the weight values provided by the author, there will be a grid of deviations, the result will be normal after add 1 in grid index. I haven't figured out why this happened. It may be that there is an error in the author's trained weight file.

xiaohu2015 commented 6 years ago

@xiaochus The problem has been found. You can ref to this issues: Keras-Yolov3.

xiaochus commented 6 years ago

@xiaohu2015 Thanks, the problem was fixed!