OIDv5?

[cesarandreslopez]

Hello!

Any plans to release a version for OIDv5 anytime soon?

[monocongo]

What are the incompatibilities with the database new version (v5)? Is there a new API?

What can I do to change the version that this tool pulls from (i.e. update from v4 to v5), in order to see any errors such a change may produce when I download images?

[keldrom]

The version 5 of the dataset is really different so the toolkit needs a big update. I promise I will do it ASAP but my time is not as huge as I need, I'm sorry

[monocongo]

So are you saying you don't want help with this (or other) issues?

[keldrom]

I will do it for sure but I need more time.

[monocongo]

Again, I'm offering to help, the above is asking for suggestions as to how to discover the incompatibilities with version 5, but your responses indicate that you're not interested in such contributions and would prefer to take on the entire job yourself. If that's the case then fine, it's your project and your prerogative, I can fork it and move forward on my own if I am so inclined, I'm just trying to get some clarification as to which way to go in case I really need to use version 5 and this project goes stagnant. I recently made contributions to fix the .gitignore and all the work necessary to add the package to PyPI but these appear to be unwelcome contributions since #52 was closed without resolution and the PR #54 is not being reviewed or merged. I'm happy to contribute to the project but only if my work is going to be useful and accepted, if not just let me know and I'll forge ahead on my own outside of the context of this project. Your call.

[keldrom]

I'm sorry! I'm pleased to have some help! If you want to try to update the toolkit the main mods are:

• [ ] change the download of the file (the URLs from v4 to v5, I don't think they've changed something else but it need some tries)
• [ ] add the Relationships dataset with all possible options
• [ ] add the Segmentation dataset with all possible options

[Rajasekhar06]

Along with the above said @keldrom I request you @monocongo to add the feature argument which can generate different annotations types like for YOLO,COCO,XML so that it will give more flexibility in creating the annotations over that huge data-set

This might Give you and IDEA!

[monocongo]

I have code that I can contribute that transforms annotations between Darknet (txt), COCO (json) and PASCAL (xml) formats. I will need to step through this code to get a handle on where that can best fit in. My suggestion is to add this as a feature request in the issues tracker.

[sliawatimena]

Where to find your code @monocongo ? I think the Segmentation dataset is more important to finish it first than the relationships dataset. Thank you very much for your help and attention.

[WyattAutomation]

I made a quick and dirty python3 script to just download the JPG files for any class or classes you want that have Segmentation Mask data from OIV5 here:

https://github.com/WyattAutomation/dirtySegmentationdownloaderOIV5/blob/master/dirtySegmentationJPGdl.py

Read ALL of the comments in this script, the full explanation of how to use it is in the comments in the script. I copied the main loop from the OIDV4_Toolkit and edited it to just grab the jpg's you want

I won't maintain or improve this, it's just how I did it. Hope it helps until this can be properly updated.

[Auth0rM0rgan]

Any news regarding OIDv5 ToolKit for segmentation Dataset (@keldrom, @monocongo)? I want to use OIDv5 Segmentation dataset instead of COCO for my approach. Also, need to convert the format to COCO Json format! any suggestions?

[monocongo]

@Auth0rM0rgan I have cooked up my own OpenImages downloader as it seemed simpler than updating this code which I found a bit hard to navigate. Feel free to add an issue there for segmentation masks support, as this is actually something I've meant to get around to. Also, I have another project for manipulating annotated datasets where I might be able to add some code to convert to COCO JSON, again feel free to add an issue or PR for that there as well. Best of luck, and let me know if I can help...