search

AlexeyAB / Yolo_mark

GUI for marking bounded boxes of objects in images for training neural network Yolo v3 and v2
https://github.com/AlexeyAB/darknet
The Unlicense
1.81k stars 681 forks source link

New Video Tagging Tool | New Tool #135

Open dexception opened 5 years ago

dexception commented 5 years ago

yolo_mark is indeed a good tool to tag images but its quite laborious.

If can simply use an object tracking algorithm on top of a video and drag an object and play/pause the video we would be easily able to create dataset. on a large scale faster.

AlexeyAB commented 5 years ago

I added o button that tracks all objects from current image to next image by using Optical Flow.

Do you mean this or something else?

Do you want more stable tracking algorithm (like cv::TrackerCSRT() (CSRDCF++) - OpenCV-contirb)?

Or do you want Kalman-filter algorithm for tracking occluded objects?

dexception commented 5 years ago

Well image based tagging takes a lot of time as you have see each image. I have 3 people sitting and tagging for a week. Big waste if you ask me.

User can draw the box after pausing the video. This way i just have to make a video and tag and write the frames. So i think this approach requires a new tool.

https://www.youtube.com/watch?v=BiLNKkfW88s

CSRT.

AlexeyAB commented 5 years ago

Currently you can slice the video into frames ./yolo_mark x64/Release/data/img cap_video test.mp4 1

Then run ./yolo_mark x64/Release/data/img x64/Release/data/train.txt x64/Release/data/obj.names Select 1 or several objects on the first frame, and then press o and hold - will be the same effect as in your video. Just try.

Yes, I will add CSRT.

dexception commented 5 years ago

@AlexeyAB Please see this video

https://www.youtube.com/watch?v=vbydG78Al8s

AlexeyAB commented 5 years ago

@dexception It looks very useful. Do you have a link to this tool?

dexception commented 5 years ago

I just saw the tool by accident as i was looking for a substitute.

sovaa commented 5 years ago

The link to the tool (from the video description on yt): https://darkpgmr.tistory.com/16