chuanenlin/drone-net

Label file format

felipecpc opened this issue · 3 comments

Quick question for the label files.

At yolo doc it says:

Where x, y, width, and height are relative to the image's width and height.

In your label files you have the real coordinates (x,y for the center of the image) and width/height, is that correct?

That is correct. :)

Got it ! So what is on the readme is not true:
The original and labeled images used for retraining can be found under the image and label folders respectively.

Before training it is needed to do the conversion, like you said on format.py
I'll try it myself :)

One question, how long does it take to train the network ?

The labels in the repo are already in YOLO format so they can be used for training directly. I see that you came from my Medium post where we used BBox annotator for labeling in Mechanical Turk - in that case, you may convert your BBox annotator labels using format.py yes.

The weights in this repo was trained for about three days on a NVIDIA Tesla from darknet53.conv.74 and you may use the weights to continue training from where I left off.