Annotation upload via CLI

Feedback
1

Hello, I’m having some trouble using the roboflow CLI.

I’ve been trying to upload a set of two thousand wild animal images, and around eight thousand images of cats and dogs. The browser was having a hard time with this, so I tried using the CLI instead. I noticed some of the import commands were commented out in the cli files, so I forked it and added this back in, and made some changes necessary for it to run on windows, and made a pull request with these. I can now get the images to upload. However, the bounding boxes seem to be gone.

The coordinates in the raw data when uploaded don’t make sense, and seem to cover the entire image rather than the area defined in the local .txt file. I’m trying to upload in YOLOv8 pytorch txt format, but I suspect the problem might stem from the fact that the CLI uses a .toVOC function to convert the files to PASCAL VOC XML format.

I’ve tested this again after downloading a dataset in PASCAL VOC format, and it works. I think this confirms that the CLI currently only works if local images are annotated in that format.

2

Thank you so much for your feedback! This has been sent over to the Product team.

We appreciate the addition of your Pull Request, too!

I do want to note, I’ve tried the CLI with some other formats, and it seems to work for COCO and YOLOv5 for me. I saw that you noted you were using Windows in your PR, so that might be the root of it. We’ll look to do some more testing to confirm