It classified my mostly-bald, short hair as "shower cap". I help a coffee cup up to the camera that it called a plunger. There were some other bathroom-centric classifications as well, maybe it tries to guess the room type and common objects from the scene? FTR, I'm sitting in my open kitchen area, nothing around looks remotely like a bathroom.
I can appreciate folks coming to the defense of the demo, but the lede on what makes this special is pretty well buried to folks that aren't in this space.
If there was a blurb on the page stating that the cool bit we should be excited about isn't whether the image classification is accurate, but that there's realtime image classification running against WebGPU. That's definitely neat. However, with no context for folks clicking a random link, I don't think it's really all that off base for folks to comment on the model's functionality being comically inaccurate. At the time of writing, I had to read the bottom comment to get the understanding necessary for why this is neat.
You should add a button to switch cameras, on my computer it’s stuck on the Open Broadcaster Studio’s virtual camera (even while close) and I’m not sure how to get it use my webcam instead.
People complaining about quality here are missing the point, this is ONNX compatible inference engine written Rust, it just using 5MB SqueezeNet from 2016 for simplicity.
Question is, is it worth to invest time and effort into ONNX ?
Missing the point?
When the classifications are horribly bad, what is the point?
I can write a random phrase generator in FAR less than 5MB that would have the same overall accuracy as this.
No, the inference accuracy of the image classifier is dependent on the model used and this is a demo of the code executing the model in a browser with GPU acceleration not the model itself. You can plug and play any model in the onnx format e.g. https://github.com/onnx/models. As a comparison, complaining about the "abysmal quality" of the dummy model on display here is like saying blender is bad 3d modeling software after opening it for the first time because all it models is a blank cube.
