Feature request: Support upload of ZIP files containing related images

[ptitzler]

As a user I might have collected a set of related images ("cats") and stored them in an archive (e.g. ZIP file). It would be nice if one could upload such an archive file and specify a label, which would be applied to all images in the archive.

Example scenario flow:

• User uploads a ZIP file
• User is prompted to choose/create labels (if no lo labels are chosen/created all images would be unlabeled)
• Archive is extracted and the selected labels applied to each image in the archive

[bourdakos1]

I'm not sure if this covers your issue, but I just added a new feature that allows you to upload a zip of prelabeled data (The only caveat is that the annotations need to be in the Cloud Annotation format)

[ptitzler]

Unfortunately not quite because it's a bit of a catch-22. If one needs to use the tool to create the annotations for the ZIP file then one would have already had to upload the images separately. While this approach might be sufficient for workshops or tutorials (where the instructor prepares the ZIP file ahead of time) it doesn't make it any easier for customers.

[bourdakos1]

Would this work?

• select a label on the sidebar
• upload a group of images (automatically labeled as the active label)

[ptitzler]

In this context, does "group of images" refer to an archive? In general it would be great if the workflow was pretty much identical for multi-selected images (user uploads a set of manually selected images) and archives (user uploads a zip file) since the user's goal can be very similar: assign one or more labels to all images in the set. Whichever way your UI design can accomplish that should be fine.

[bourdakos1]

I think I want to leave uploading an archive restricted to pre-annotated images

I think it could cause some confusion mixing the 2, unless you have a good reason why uploading a zip file full of images has an large enough advantage over uploading a group of selected images

[bourdakos1]

I think it might also add a lot of code overhead, especially since I'm hoping to add support for importing additional annotation formats

[ptitzler]

I think I want to leave uploading an archive restricted to pre-annotated images

That would be sufficient for demos/workshops where somebody wants to use canned data set. (more below)

I think it could cause some confusion mixing the 2, unless you have a good reason why uploading a zip file full of images has an large enough advantage over uploading a group of selected images
Primarily re-usability and (ideally) a faster data transfer.

Maybe a meet-in-the middle - make the embedded annotation optional if it is not already? In essence a user can upload a ZIP file that only includes images (no auto-annotation is performed) or a ZIP file that contains images and annotations (auto-annotation is performed)

[bourdakos1]

At the moment, if they upload a zip file with no annotations, it still uploads the images in the zip as unlabeled.

prompt after upload:

I don't really want to prompt the users to choose a label when upload a zip file. Personally, I get a little stressed when I get asked a question as the result of uploading a file. I want to avoid Modals/prompts because I think they get a bit intrusive in these situations

pre-choosing a label before upload:

If for some reason they upload a zip with annotations, but the tool doesn't see the annotations, it will try to dump all the images in whatever label happens to be open. This probably won't happen often, but with this UI interaction I don't think it's clear enough what's going to happen when uploading a zip.

At the end of the day, the tool's purpose is annotating images. If your images are already organized by folders, you don't really need to annotate them haha 🤷‍♂️.

I think the root issue is just the need for an easy way to train a model with images that are already organized into folders? Maybe another solution is extending cacli, so you can train by pointing at the folders?

(ideally) a faster data transfer.

(I have to extract all the images before uploading them to COS)