My second pose model, and I think this one turned out better now that I've learned what not to do. This one was generated with a Google Colab program.
If you're more of an expert than I am on this, feel free to peer into the code and let me know what I could have done better.
Description
First version.
FAQ
Comments (3)
A quick look at your dataset some things jump out; 1. a few duplicates that can be removed
2. a couple images have more than one person, consider removing those.
3. crop the images that have text.
Those I would consider to be the majors.
Another consideration would be to possibly divide the sets into arm classifications as well (grand jete + arms fifth position, grand jete + arms fourth position, grand jete + arms third position, etc).
Thanks. I was wondering why so many of the images were zoomed out.
@zorpfurble something like grand jete I would personally want to have both zoomed out and closeup. The zoomout adds that flare of the leap in contrast to the staging area, but definitely a loss of detail. If you really wanted to get nitpicky about it because you're training for pony is to make sure the shortest side is 1024, and remove any images that would be crap quality at that size.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.