Clair Obscur: an Illumination-Aware Method for Real-World Image Vectorization (CVPR 2026 Highlight)

arXiv License

Xingyue Lin, Shuai Peng, Xiangyu Xie, Jianhua Zhu, Yuxuan Zhou, Liangcai Gao
Wangxuan Institute of Computer Technology, Peking University

---

Description

Official implementation of Clair Obscur: an Illumination-Aware Method for Real-World Image Vectorization.

COVec is an illumination-aware vectorization framework inspired by the Clair-Obscur principle of light-shade contrast.
It introduces intrinsic decomposition in the vector domain, representing each image with albedo, shade, and light layers in one editable SVG.

Progressive composition from albedo, shade, and light; editing by modifying albedo while preserving illumination.

The principle of Clair-Obscur in art. Classical painting and modern animation use tone variations within the same semantic regions (e.g. skin, hair) to convey light–shade structure.

Code

Requirements

• Python 3.10+ with CUDA 12.x GPU
• pydiffvg (built from source)
• SAM ViT-H checkpoint
• Intrinsic (optional, for --generate_albedo)
• Stable Diffusion v1.5 (optional, for SDS simplification)

Installation

Quick install

conda create -n covec python=3.10 -y
conda activate covec
cd Pipeline
chmod +x setup_single_env.sh
./setup_single_env.sh

Manual install (order matters, pin torch 2.4+cu124)

pip install torch==2.4.0 torchvision==0.19.0 \
  --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements-base.txt
bash install_diffvg.sh
pip install -c constraints.txt git+https://github.com/compphoto/Intrinsic.git

Download SAM ViT-H to Pipeline/checkpoints/
(see Pipeline/checkpoints/README.md)

Usage

Run from Pipeline/:

# Full pipeline with automatic albedo generation
python pipeline.py \
  --run_all --generate_albedo --image_name 2-thing-2.png --path_num 64

# Full pipeline (albedo already prepared)
python pipeline.py \
  --run_all --image_name 2-thing-2.png --path_num 64

Place inputs under target_imgs/init/ and albedo references under target_imgs/albedo/.
Output: workdir/<image_name>/<path_num>_paths/result.svg

Generate albedo only

python utils/albedo_generator.py \
  --input ./target_imgs/init/2-thing-2.png \
  --output ./target_imgs/albedo/2-thing-2.png

More qualitative results: examples

Notes & troubleshooting

Topic	Detail
PyTorch	Pin 2.4+cu124; install Intrinsic with constraints.txt
pydiffvg	Build with bash install_diffvg.sh only
GPU memory	release_intrinsic_after_albedo: true frees VRAM before SAM
Intrinsic	Academic use only - see license

If pydiffvg build fails (CUDA / GLIBCXX mismatch), set CUDA_HOME=/usr/local/cuda-12.4 and rerun bash install_diffvg.sh.

License

This project is released under the Apache 2.0 License.

Third-party: pydiffvg (Apache 2.0), SAM (Apache 2.0), Stable Diffusion v1.5 (CreativeML Open RAIL-M), Intrinsic (academic only).