I'm looking for some offline software that can be used for labelling many images with rectangular bounding boxes for the purposes of object detection. It needs to be able to import pre-generated labels which can then be validated / tweaked by a human and then saved. Ideally it would work in YOLO format. I have found that: VoTT doesn't support importing labels (only importing CNTK models) labelImg is buggy - and at least for my dataset unworkable Yolo_mark is awkward and unproductive - requires several clicks for every annotation.
2 Answers
There is a list of annotation tools that seems to be relatively well maintained here: https://www.datasetlist.com/tools/
I haven't tested them all, but my favourites so far are
- deeplabel (has exactly the features I need, easy to use)
- VIAME (very powerful, more difficult to use because it has so many options)
- scalabel (excellent look and feel, good team behind it, still a few bugs to iron out)
I agree with OP that yolo_mark and VoTT didn't live up to my expectations.