thtrieu / darkflow

Translate darknet to tensorflow. Load trained weights, retrain/fine-tune using tensorflow, export constant graph def to mobile devices
GNU General Public License v3.0
6.13k stars 2.08k forks source link

Why the ground truth is the absolute coordinates instead of the ratios #1060

Open mylife126 opened 5 years ago

mylife126 commented 5 years ago

Hi,

I am really new to YOLO. However, as I am looking at the XML file, I found that the ground truth of the bounding boxes are the absolute coordinates for each boxes. However, as I read the paper, should not the ground truth to be {bx,by,bw,bh}, and bx by represent the aspect ratio of where the object is with respect to its located grid cell, which is bounded between 0~1 ?

I am asking this because I want to create my own dataset, and at this point, I do not know how I should annotate my ground truth of each the bbox's coordinate.

Thanks for helping me,

Regards.