How to run a pytorch model on the browser?

Question

I would like to know what are the options I have to run an object detection model in the browser, until now I have found the next options:

1. Streamlit: is very simple but requires the server resources to run, including its own camera. Besides, I believe is not possible to be implemented as a website.
2. Brython: is a JS substitute? I couldnt find the way how to load pytorch or libraries.
3. Onnx: Seems to be a very good option, however I dont quite understand how to run the model in JS, since I have been using python and pytorch only, What is the learning roadmap? What is the relation with Webassembly?
4. NCNN: This is the method I know the least, It seems to be able to work on basically any device, Webassembly, windows, android..., but again, I cant understand how to use my python-pytorch model to this framework since I dont know the roadmap.

Are there other options?

Motion

Answer 1

There's TensorFlow.js.

You still need to convert your model with something like ONNX, though.

Pallor

Answer 2

Just to update you. Tensorflow.js is suitable for the TensorFlow model while onnx runtime web is a good option for others. Things to do are;

• Convert your model to onnx(see PyTorch export onnx)
• Use onnxruntime web to use generated onnx with javascript(detailed at onnxruntime.ai)

If anyone knows more options, kindly add.

Timber

Answer 3

Another alternative is JS-PyTorch.

You can train and run models directly on the web (with GPU support).

Nutritious

While this link may answer the question, it is better to include the essential parts of the answer here and provide the link for reference. Link-only answers can become invalid if the linked page changes. - From Review – Olatha