tfgo VS CloudForest

eh, awesome! Seems this one, right? https://github.com/galeone/tfgo. Quite many stars.

I'm excited to share a project I've been working on: [GoFaceRec](https://github.com/modanesh/GoFaceRec). This is a face recognition tool built in Go, leveraging the power of MTCNN for face detection and QMagFace for face recognition. The project was born out of a desire to bring the power of deep learning models to the Go community. After much effort, I concluded that the best approach was to convert models to TensorFlow and then work with tfgo, a Go binding to TensorFlow's C API. In GoFaceRec, the input image is first processed, and then its embeddings are compared against the ones already computed from our dataset. If the distance between embeddings falls below a specific threshold, then the face is considered as unknown. Otherwise, the proper label will be printed. The project is tested using Go 1.17 on Ubuntu 20.04. For gocv, the version of OpenCV installed is 4.7. And for tfgo, I installed [this version](https://github.com/galeone/tfgo) instead of the official one. You can install this package by running the following command in your project: > go get github.com/modanesh/[email protected] You can find more detailed instructions on how to use the tool in the [GitHub repository](https://github.com/modanesh/GoFaceRec). I welcome any feedback, suggestions, or contributions to the project. I'm looking forward to seeing how the community uses GoFaceRec and hope it can be a valuable tool for those working on face recognition tasks. Happy coding! 🚀

Paolo Galeone has improved bindings (tfgo) that can be used for training and deployment.

https://github.com/galeone/tfgo here is a very good tutorial. I would suggest starting there.

Tensorflow is actually written in C++, and the python package is just bindings to tensorflow. There are Tensorflow Go bindings: https://github.com/galeone/tfgo.

I implemented something like this in a [pre xgboost boosting framework](https://github.com/ryanbressler/CloudForest) ~10 years ago and it worked well.

It isn't even that much of a speed hit using the classical sorting CART implementation. However xgboost and ligthgbm use histogram based approximate sorting which might be harder to adapt in a performant way. And certainly the code will be a lot messier.

I've got a ~10 year old implementation that does something similar calling it "three way splitting" here: https://github.com/ryanbressler/CloudForest

And i got the idea from a lab mate, Timo Erkkila's RF-ACE project though neither of us thought it was a particularly novel idea.

The best method i've seen for dealing with this bias is to create "artificial contrasts" by including possibly many permutated copies of each feature and then doing a statistical test of the random forest importance values for each feature vs its shuffled contrasts. This method is described here: https://www.jmlr.org/papers/volume10/tuv09a/tuv09a.pdf and there is an implementation here: https://github.com/ryanbressler/CloudForest