kylemd / gcam-autotune

Using Frida to patch GCam libs live in RAM, retrieve photo and assess result

GCam-Autotune

Auto tuner for finding the optimum values for Google Camera lib tunables.

How does it work?

We iteratively test the effects that changing each parameter has on image quality.

We use Facebooks' ax-platform package for Bayesian optimisation to try and find the optimum value in as few tests as possible - we don't want to be running 100,000s of experiments on each value as it would take forever!

Does it work?

This is a very experimental - it is my first Python project so expect bugs!

I upload everything I do so people can collaborate and I can learn. Don't be shy - fork, modify, test, post issues etc.

Requirements

A Google Camera mod installed.

Root on the handset you want to test.

Python 3.10.8 and git installed.

Pytorch set up on your machine

PyCharm or VS Code (recommended, not needed - helps with debugging and reporting issues)

How to use

I recommend using pyenv to install the required Python version.

Use git clone to get this repository onto your hard drive and open it up in PyCharm.

Once you've done that, run pip install -r requirements.txt in the project root to install all dependencies.

Go into AutoTuner.py and make sure the values in args_dict are correct. If you want to run the tests proper, remove the testParam field - this is only one tunable that I'm using for debugging currently.

To-do:

More robust exception catching

Correctly read user variables from JSON

De-duplicate the values in Rivovs' API

Write code to roll-back changes to lib if that tunable isn't working