Cvat: Add/Delete images/video to existing task

[carlescarmonacalpe]

Hi,

Thank you for this amazing tools, seems really promising. I'd like to suggest a new feature which is to allow to add/delete images/videos from an existing task. From my side it's really common to extend or reduce the dataset size during the development process.

Thank you for you work!

[nmanovic]

Hi @carlescarmonacalpe ,

Why is it not possible just to create one more annotation task with new images? What is the advantage to update the same task?

[carlescarmonacalpe]

Hi,

The only advantage is that you don't need to modify/join the exported files If you annotate two different tasks and you need to have both annotations together to train a model.

[nmanovic]

Hi @carlescarmonacalpe ,

If you have a good idea how to implement that please send us a PR. Our internal focus is to add stability, introduce analytics, improve UI and implement a couple of semi-automated methods to annotate data. Please let me close the issue for now but if you like implement such functionality don't hesitate to reopen it. In any case I really appreciate your involvements. Thank you very much!

[nmanovic]

Probably it isn't a bad idea to treat an annotation task like a dataset (which can be extended or reduced). It is especially good if we will be able to change status of an annotation task automatically (e.g. done -> annotate). Also a couple of internal customers also requested the feature.

[nmanovic]

Vitor Casadei 04:41
Hi, I used to use vatic and it exported the videos in image format. However on CVAT I can only download the labels, is that right? I've got to convert frames to images on my own?

vcasadei , at the moment you need to prepare data (at least for videos) by yourself. Internally we don't keep original data to save server disk space. When you upload data CVAT does compress images (image_quality parameter when you create an annotation task). Thus 1 frame can weight only 10-100 kilobytes. For example, an original image in png format can be several megabytes. I will submit an issue regarding that. Probably we need one more step in CVAT: prepare a dataset. Thus users can prepare data and after that download them from CVAT as is. If you can help us with such app/feature for CVAT you are welcome.

[nmanovic]

The feature will be implemented only in our new UI. Thus need to move it on v1.0.0 which is expected at the end of the year.

[pagujeeva]

I am facing similar issue where I will have to consolidate/add additional frames with annotation and generate one single JSON file. Is there a way, I can do that?

[illuminati945]

Hi,
Is there a progress update on this feature?

[michellemul]

Hello! Is there anything more with this? I have an extra 40 images to annotate so was hoping to just add them into my existing annotations. Cheers!

[mrawding]

Hi, any update on this feature? Would be extremely useful !! CVAT is great, but this definitely helps users who might have made mistakes when uploading datasets and didn't realize until after annotation. Furthermore, it is beneficial to any user who would like to continue to add to their datasets as they annotate. It makes exportation of annotations easier and more consolidated when going to train.

[nmanovic]

@mrawding , the PR is the first step to solve the problem https://github.com/openvinotoolkit/cvat/pull/2255