Strand Integration (PG2023)

Refinement of Hair Geometry by Strand Integration,
Ryota Maeda, Kenshi Takayama, Takafumi Taketomi,
Computer Graphics Forum (PG2023)

This repository contains the following implementations:

• Strand Integration: The official implementation of our paper "Refinement of Hair Geometry by Strand Integration" (PG2023).
• LPMVS: The unofficial implementation of LPMVS (Line-based PatchMatch MVS) proposed in the paper "Strand-accurate Multi-view Hair Capture" (CVPR2019). And we also provide the implementation of 3D line filtering.

Multi-view data

Small-scale data

This repository contains small-scale multi-view images of straight hair for debugging our code. This data is identical to the data used in our paper except that it is resized to 1/10. You can see it in data/straight_s.

Full-scale data

You can download the full-scale multi-view images from release page.

Setup

Our code requires to install the our custom Python extension implemented in C++. Please build and install it before running our code.

cd cpp_ext
pip install .

For more details, please refer to cpp_ext.

Running

# LPMVS
python run_lpmvs.py data/straight_s -o result/lpmvs/straight_s 
# Estimate a 3D line map for each view from multi-view images.

# Run 3D line filtering (for LPMVS)
python run_line_filtering.py result/lpmvs/straight_s result/merged_ply/lpmvs/straight_s.ply
# Merge the 3D line maps into a single point cloud file.

# 3D line consistency map
python run_consistency_map.py result/lpmvs/straight_s -o result/consistency/straight_s
# Generate a 3D line consistency map described in Subsection 3.5 in our paper.

# Strand Integration
python run_strand_integration.py result/lpmvs/straight_s --consistency result/consistency/straight_s -o result/si/straight_s --views 5 --imshow
# This will only apply for view 5. You can run for all views by removing the --views option.

# 3D line filtering (for Strand Integration)
python run_line_filtering.py result/si/straight_s result/merged_ply/si/straight_s.ply
# After running all views, you can merge the results same as the above process.

After running the reconstruction codes, you can visualize the .ply files as follows.

# Visualize ply files
python visualize_ply.py result/merged_ply/si/straight_s.ply --gamma 0.45

BibTeX

@article{maeda2023refinement,
    author = {Maeda, Ryota and Takayama, Kenshi and Taketomi, Takafumi},
    title = {Refinement of Hair Geometry by Strand Integration},
    journal = {Computer Graphics Forum (proceedings of Pacific Graphics)},
    volume = {42},
    number = {7},
    year = {2023}
}