A live demo of MIRNet-TFJS

[Rishit Dagli]

Hi,

A live demo of the MIRNet-TFJS repo, capable of enhancing low-light images to great extents is now out for you to try out.

Try it: https://mirnet-tfjs.rishit.tech

Project repo: https://github.com/Rishit-dagli/MIRNet-TFJS

More details in this tweet.

[Jason Mayes]

This is awesome work Rishit! Thank you for contributing to the community and making such epic models available in TFJS (even if via Node for now until the model size comes down for a client side deploy) :-)

If anyone has any thoughts on how to shrink this model for better browser consumption do reply :-)

Cheers!

Jason

--
You received this message because you are subscribed to the Google Groups "TensorFlow.js Discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tfjs+uns...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/tfjs/23f9aa83-5d9c-4ff8-8dfc-c318bce67dc6n%40tensorflow.org.

--

•  Jason Mayes   •  Senior Developer Advocate, TensorFlow.js   •  Google   •   Follow me or get inspired with #MadeWithTFJS demos

This email may be confidential and privileged. If you received this communication by mistake, please don't forward it to anyone else, please erase all copies and attachments, and please let me know that it has gone to the wrong person.

[Sayak Paul]

One immediate thing to be done is to compress the original model even further with distillation. 

This is on my radar since I worked on the TFLite variant of this same model: https://tfhub.dev/sayakpaul/lite-model/mirnet-fixed/dr/1. 

Some more context as to why distillation could be really helpful: in this notebook you will find 43 KB super resolution model distilled from a much larger model: 

https://github.com/sayakpaul/Adventures-in-TensorFlow-Lite/blob/master/ESRGAN_TFLite.ipynb. 

Sayak Paul | sayak.dev

To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/tfjs/CAK49QddK3VBjA35baGN%2Bo_9XCTADnnoaAr2z5bzvOaHOpHtyqQ%40mail.gmail.com.

[Jason Mayes]

Wow that is quite the compression. Would love to see how fast this could potentially run in browsers with the optimizations you were using? Imagine using this in a browser to enhance low light webcam streams in real time. That would be epic.