-
Notifications
You must be signed in to change notification settings - Fork 2.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
How to keep track id? #21
Comments
Hi @abcilike , At the moment it isn't possible but we will consider to resolve the problem in one of our next releases. Thank you for your question. |
I found that the annotation file can save the target ID when importing the video, but not when importing the image sequence. |
It doesn't work for videos in general case. The main problem here that each time when you press "save" button annotations are written to DB from scratch (not an increment). Need to track ID explicitly as an automatic attribute. |
Hello, @nmanovic |
@SunskyF , You should create "interpolation" task for that. It is created automatically if you upload a video file. If you load images the task will be considered as "annotation". At the moment the only way to dump "tracks" for "annotation" tasks is to change DB manually or change code (https://github.com/opencv/cvat/blob/master/cvat/apps/engine/annotation.py#L1015). Does it answer your question? |
@nmanovic |
#120 implements client_id for each annotation object. |
Is this feature already in develop branch? If I give image sequences as input the dumped annotation file still does not contain track ids. Instead, if I input the video file, the annotation file does contain the track ids. In both cases I made sure I was annotating in interpolation mode. I wonder if I am doing something wrong --- why track ids are not dumped for image sequences? |
@sjoycode , at the moment if you annotate images you annotation will be in "annotation" format: https://github.com/opencv/cvat/blob/develop/cvat/apps/documentation/xml_format.md#annotation |
Are there still plans to integrate an object ID into the annotation format anytime soon? For tracking purposes this is crucial and a lot of labelling effort is lost if not being able to save the object id for later use. Especially when labelling frames of a video as single images using the interpolation mode. |
Hi @sttobia , first of all you can workaround the issue. Go to admin panel and change "mode" for a task from "annotation" to "interpolation". You will loose image names but you will be able to dump the task in "interpolation" format. In our nearst plans is to improve export/import annotations in different data annotation formats. It will close the issue completely. You will be able to upload/download annotations in most popular formats like COCO, PascalVOC, etc... I realized that it is the most critical feature for many users. |
Thanks for the quick reply! Does this also work for images and not videos? I am using single frames named using a sensor timestamp which is crucial for later reuse of the labels. Therefore the image name is somehow also critical for me. |
Hi, am wondering is this issue fixed? |
Hello, have you found an answer to this? |
Hi, has this label upload improvement been implemented? I am still experiencing the issue of tracking labels turning back to regular labels upon upload of labels in COCO 1.0 format. This feature is a critical requirement for my project. I look forward to hearing back from you. Thank you. |
I'm in the same situation. I have been able to dump ids for objects manually drawn when creating task from video, but I am not able to import annotations with the ids, it rewrites them. It's not a problem at all the change of ids itself, the biggest problem is that it uses new ones for each frame. So it completely ignores the tracking part. I have tried with both MOT and PascalVOC format |
I don't mind trying to modify the function in charge of it and doing a pull request if I make it. But I do not longer see the file cvat/apps/engine/annotation.py |
After annotation done --> upload annotation:
The ID of the target is messed up.
How to include the id information of the calibration target in the exported annotation file, and re-import the annotation file without showing confusion.
The text was updated successfully, but these errors were encountered: