cvat-ai / cvat

Annotate better with CVAT, the industry-leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale.
https://cvat.ai
MIT License
12.73k stars 3.02k forks source link

Remove the labeling rectangle in last frame from a video file cause all previous labeling rectangles go away #2999

Open shaojun opened 3 years ago

shaojun commented 3 years ago

Expected Behaviour

the last several frames does not contain the target object, so I select the labeling rectangle and delete it, the previous frames with target object SHOULD still keep the labeling rectangles. Or at least prompt to warn user this operation will clear all previous labeling rectangles.

Current Behaviour

the labeling rectangles in previous frames are all gone.

Your Environment

Server version: 1.2 Core version: 3.10.0 Canvas version: 2.2.1 UI version: 1.13.3

This is a disaster for me that hundreds of previous labelled frames are all lost labeling rectangles, without any prompt.

nmanovic commented 3 years ago

@shaojun , probably you are right. For tracks the delete operation should mark a frame as outside by default.

shaojun commented 3 years ago

@nmanovic thanks for this great repo and quick response here, if this behavior is confirmed, then I'm personally think this is a high priority bug rather than a enhancement, reason:

nmanovic commented 3 years ago

@shaojun , the current behavior was for years. Users can always undo operations if they did it by an accident. But I already heard several complains about the behaviours.

shaojun commented 3 years ago

@nmanovic this is not by user accident, it's on purpose operation since some frames may not contain the target object.

Could you suggest for current version, how to remove labeling rectangles for target-object-not-exists frames?

nmanovic commented 3 years ago

@shaojun , if my guess is correct and you are annotating a video file, you need to mark the rectangle as outside. See documentation for more details: https://github.com/openvinotoolkit/cvat/blob/develop/cvat/apps/documentation/user_guide.md

shaojun commented 3 years ago

@nmanovic thanks for the point, what I'm doing is uploading a video file which taken by my phone and then try labeling all target objects in each frames and finally export to VOC format (each frame will be exported as a JPEG file) for object detection training purpose.
(I take the video rather than pictures is because I think taking phone pictures is slow)

please help to confirm, with the track mode, when the object disappeared in the middle of serials of frames, I should click the switch outside property to dis-continue the track rectangle shown from previous frames, and then repeat Go Next until the frame of object show again, then I should draw another track, is this correct way?

zhiltsov-max commented 3 years ago

@shaojun, yes, outside works exactly the way you describe. You'll probably need to enable display of invisible annotations in the UI settings to select and continue the track when the object reappears in the video.

shaojun commented 3 years ago

@zhiltsov-max thanks. could u clarify a bit about the purpose of the merge track, as i think the 3 steps: draw track-> outside property -> redraw another track already cover case.

zhiltsov-max commented 3 years ago

To support an other way to create a single track. There is also a split track button nearby.