onnxruntime-web vs tfjs

[MariasStory]

Issue Description
tfjs models are slower then onnxruntime-web

Steps to Reproduce
https://github.com/microsoft/onnxruntime-web-benchmark

Expected Behavior
Use the fastest framework for a specific task

Environment
web

[vladmandic]

this is a veeeery generic statement without any provided information...

• which exact models?
• using which backend?
• what is the performance difference - give specifics

[MariasStory]

Hi Vladimir,
Thanks for you comment. You are doing a really good job. I really like your solution.
I am sorry for not being specific.
In my test, I was using a simple image processing with models from https://github.com/PINTO0309/PINTO_model_zoo
There are different formats of the models, so it is easy to compare.
When using onnx models I have seen smaller memory footprint and faster execution.
There is a chance that it is a specifics of the models conversion, in my specific case.
I just want to suggest to try different approach. The onnx implementation is relatively easy to implement and there are plenty models to use.
In my case, I just used simplest single html implementation with:
<script src="https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/ort.min.js"></script>

[vladmandic]

it really depends on the model and the backend used
different models use different kernel ops and those ops are implemented differently by onnx and tjfs

even more, they are very different comparing tensorflow backend (in nodejs) or wasm or webgl or webgpu.
as far as i know:

• onnx really prefers wasm backend and support for webgl is quite limited
• tfjs doesn't spend that much time optimizing wasm backend, its the least one preferred and much more development is focused towards webgl (current) and webgpu (future)

that's why i'm asking - which exact model and which exact backend?

[vladmandic]

btw, as this is not really an issue, i'm converting it to a discussion item.