We provide a slightly modified version of the evaluation code from StyleLight and Editable Indoor LightEstimation. While we adapted the code to better suit our needs, the underlying evaluation method remains the same. You may use either this repository or the original code; both should produce identical score outputs.
This directory contains an example predicted output and an example ground truth in the example directory.
cd stylelight
python envmap2resize.py --input_dir ../example/hdr/stylelight/ --output_dir ../output/stylelight/envmap_128x256
python tonemap.py --testdata ../output/stylelight/envmap_128x256 --out_dir ../output/stylelight/envmap_toned
python job_distributor.py --input_dir ../output/stylelight/envmap_toned --output_dir ../output/stylelight/rendered
sh test_rmse.sh ../output/stylelight/rendered ../example/ground_truth/stylelightThe score will be printed to the terminal and saved to a CSV file at . ../output/stylelight/rendered.
Even though both DiffusionLight and StyleLight are outputs at 256x512, we still run envmap2resize.py to resize the environment map to 128x256 to match the StyleLight evaluation method described in the implementation details of the StyleLight paper.
cd editableindoor
python job_distributor.py --input_dir ../example/hdr/editableindoor/ --output_dir ../output/editableindoor/hdr --output_ldr ../output/editableindoor/ldr
python res1_table.py --input_dir ../output/editableindoor/hdr --output_dir ../output/editableindoor --gt_dir ../example/ground_truth/editableindoorThe score will be printed to the terminal and saved to a CSV file at ../output/editableindoor
cd envmapnet
python compute_directional_error.py --input_dir ../example/hdr/envmapnet/ --gt_dir ../example/ground_truth/envmapnet/ --output_dir ../output/envmapnet The directional (angular) error is taken from EnvMapNet
The directional error will be printed to the terminal and saved to a text file at ../output/envmapnet in the degree unit and also produce csv file in radiant unit
Please follow the instructions in our main repository's installation guide
Install Blender on your machine. At the time of writing this paper, we were using Blender version 3.6.5. However, any version of Blender newer than 3.0 should work.
@inproceedings{Phongthawee2023DiffusionLight,
author = {Phongthawee, Pakkapon and Chinchuthakun, Worameth and Sinsunthithet, Nontaphat and Raj, Amit and Jampani, Varun and Khungurn, Pramook and Suwajanakorn, Supasorn},
title = {DiffusionLight: Light Probes for Free by Painting a Chrome Ball},
booktitle = {ArXiv},
year = {2023},
}
