Add instant annotation donwload! How to deal with growing datasets?

[Shuai-Xie]

Let's see this case.

A pipeline has been defined.

But if

• The images in my dataset folder are expanding, but the pipeline annotation progress will not.
• A new class needs to be introduced, but the LabelTree of the pipeline can't be updated.
• I just want to export some already labeled data annotation, but the pipeline makes me have to finish all the annotation task before I export the annotation.

The pipeline is a good character, however, it needs more flexibility.

[jaeger-j]

Hi @Shuai-Xie, thank you for your contribution.

The images in my dataset folder are expanding, but the pipeline annotation progress will not.

Please use an iterative pipeline to solve this problem. If your dataset is growing, while a pipeline has already been started, you can use loop elements. An possible pipeline flow could be as follows:

1. Select 100 images from your dataset.
2. Perform an annotation task for theses 100 images.
3. A loop elements jumps back to step 1.

A new class needs to be introduced, but the LabelTree of the pipeline can't be updated.

Please see my answer to issue #40

I just want to export some already labeled data annotation, but the pipeline makes me have to finish all the annotation task before I export the annotation.

That is a nice point. We should add a button to each annotation task in our pipeline view for an instant download of the annotations.

[Shuai-Xie]

Hi @jaeger-j Thanks for your detailed reply. I'll try your suggestions right away.

So fast does the GitHub stars of LOST increase, That's indeed a promising work.

[avi-vsionai]

Has there been any development on this issue? I would love to download partial annotations without having completed the entire dataset. For the time being I'll also take any dirty hack.

[jaeger-j]

Hi @avi-vsionai,
Please use our instant export pipeline for now. In LOST 2.0 there will be a GUI-button.

[gereonreus]

Already implemented on branch 2.x