Contribution

[Autumn-Ou]

During the 2023 season I developed a heatmap based key pointer built on a U-net MobileNet architecture that can be easily adapted by changing the dataset & number of key points and I'm betting your team could make better use of it than most given your use of jetson coprocessors (and pytorch models inability to run on edgetpus) and I was wondering if you'd be interested in adding it to your workspace, I'm aware that it's not based in tensorflow and might not be a good fit but I thought your team may be interested in having it since it is built to run easily on CUDA (or ROCm) supporting devices

An example of it in use with only a 150 images of training data