lightly-ai
First comit to add video instance segmentation
Hi, Thank you for your clean and well-organized code! I'm adding a new feature to support video instance segmentation, as it's necessary for one of my projects, and I would be honored to contribute to this project. My goals are:
- Add conversion between VIS datasets like YouTube VIS, KITTI MOTS, and others.
- Add conversion from VIS datasets to image datasets.
- Follow the general principles of your code as well as writing unit and integration tests. You can view the current status in my forked repository. If you believe I can merge this into the main repository in the future, would you create a new branch for "VIS"?
Hey @fardinayar, thank you for the PR. We will have a look at this asap. Do you have any suggestion for a small video dataset to use to test the conversion outside of the added tests?
Hi again, Thanks for giving it a chance! Actually, I haven't tested it except with datasets in the YouTubeVIS format, but I plan to test it with the KITTI MOTS dataset, and I will send you a small version of both datasets. For now, I'm still working on it, and there are a few things to do before it's fully ready. In the meantime, I would appreciate any suggestions in implementations.
The main reason for this early PR is to check if the idea aligns with your repo and if it’s something you'd consider.
Thank you for your detailed feedback. I will make sure to apply all of them and get back to you soon.