search

infinitered / ai-lab

Library of components for TensorFlow.js in web frameworks.
MIT License
46 stars 7 forks source link

[FEATURE] Add PoseDetection model to AILabImage #44

Closed GantMan closed 2 years ago

GantMan commented 2 years ago

Existing:

Currently, you can get an AI image component and and give it an SSD model or a classification model. When you give it a classification model, it doesn't draw anything, it just gives you the data onInference.

Task

Update AILabImage to take a third model "pose", and in the modelInfo, do a proof of concept that it can take blazepose.

Nothing needs to be drawn to the screen, but just like a classification, the results should pop out onInference so we can use it.

Q&A

Q: Why only AILabImage? A: Because, at the time of this ticket, that is the only component that supports two models.

Q: Just blazepose? Or should it support movenet, too? A: I'd love for it to support movenet, too, but that should be a stretch goal. Blazepose is the current goal.

Q: What API wraps the pose models? A: It's pose-detection (https://github.com/tensorflow/tfjs-models/tree/master/pose-detection)

GantMan commented 2 years ago

@shivaylamba - can you take this one?

GantMan commented 2 years ago

This is solved in this branch: https://github.com/infinitered/ai-lab/tree/pose_add

I'll do a PR when I'm done testing it.