Sillystill - Recreate the look of Cinestill-800T using Deep Learning

While halation and chromatic aberrations are sometimes considered as defects, artists often use these optical flaws for aesthetic purposes. Most films contain an anti-halation filter, but the Cinestill 800T does not. This produces a beautiful red halo around light sources, and has a Tungsten white balance, which creates a colour contrast between red and green tones. In this project, the goal is to recreate digitally the look of this specific film stock focusing on three aspects:

the film grain: this film stock has a high sensitivity and produces a lot of grain, which is also pleasing aesthetically
the halo: the halo is the main characteristic of this film stock
the colour profile: the colour profile is not the one of your digital camera.

This project trains various image-to-image deep learning models on a novel dataset of image pairs capturing the same scene on a digital and film camera.

🔗 Shortcuts

Here is a list of things that you likely want to do:

Check out the demo of the model (Not yet available)
Find all project details in the full report (Not yet available)
Inspect the experiment logs on W&B (Not yet available)
Download the image pair dataset from Zenodo or Huggingface (Not yet available)

⚙️ Installation

Pip

# clone project
git clone https://github.com/mikasenghaas/sillystill
cd sillystill

# [OPTIONAL] create conda environment
conda create -n sillystill python=3.9
conda activate sillystill

# install pytorch according to instructions
# https://pytorch.org/get-started/

# install requirements
pip install -r requirements.txt

Conda

# clone project
git clone https://github.com/mikasenghaas/sillystill
cd stillystill

# create conda environment and install dependencies
conda env create -f environment.yaml -n sillystill

# activate conda environment
conda activate sillystill

Pre-Commit

Before contributing, please make sure that you install the Git hooks of pre-commit (e.g. it will be triggered on every git commit) by running the following command:

pre-commit install

To check if everything works as expected you can run pre-commit run --all-files which will run all hooks on the entire repository.

🤖 How to run

Train model with default configuration

# train on CPU
python src/train.py trainer=cpu

# train on GPU
python src/train.py trainer=gpu

🙌 Acknowledgements

This work was kindly funded by the Image and Visual Representation (IVLR) Lab at EPFL as part of the Computational Photography (CS-413) class and supervised by Raphaël Wan-Li Marc Achddou and Sabine Süsstrunk. Thank you for making this project possible!

This repository is bootstrapped from this Lightning & Hydra Template.