Illumination Distribution Estimation

This is the official repository of

Physically-Plausible Illumination Distribution Estimation, Egor Ershov, Vasily Tesalin, Ivan Ermakov, Michael S. Brown. ICCV 2023. Link.

Setup

For Linux

python -m venv venv
source ./venv/bin/activate
pip install -r requirements.txt

For Windows

python -m venv venv
./venv/bin/activate.bat
pip install -r requirements.txt

Dataset

In this dataset we have provided new markup for mirror ball for Cube++ dataset. All markup is provided in markup folder markup.

Dataset generation

To generate GT distributions we need to use script data_generator.py from folder markup_generation. This script is to generate ball .PNG crops, ball masks for generated .PNG crops, ball markup w\ maximum level on the full image for every channel using full image .PNGs and ball markup for cube crops, as well as GT illumination distribution .PNG and .npy files for mirror ball crops using crops .PNGs and masks.

The usage example with local paths:

python dataset_generators/dataset_generator.py --path_to_markup markup/ --path_to_save_cube3p ../data/cube_3p/ball_crops_png/ --path_to_cube2p ../Cube++/

Solutions

#TODO