3DSSL-SS23_HairReconstruction

This repository contains the implementation of the paper: Giljoo Nam, Chenglei Wu, Min Kim, and Yaser Sheikh. Strand-accurate multi-view hair capture. pages 155–164, 06, 2019.

Note that the last part of the pipeline suggested in the paper - Hair Growing, is not implemented.

This implementation was based on the repositories:

• gipuma
• fuisible

The code in the folders /gipuma and /fuisible was modified to perform Line-based PatchMach Multi-view Stereo as proposed by the Strand-accurate multi-view hair capture paper.

Generate Orientation Fields and Confidence Values

Setup

• Ensure you have OpenCV installed on your system.
• Navigate to the Hair_Orient2D directory.
• Create a build directory and compile the program using CMake:

mkdir build
cd build
cmake ..
make

Usage

Run the program by providing the path to the input image folder:

./Orient2D <input_image_folder>

Replace <input_image_folder> with the path to the folder containing the images you want to process.

Running Line-based PatchMatch MVS

Some important files:

• gipuma/gipuma.cu: File containing cuda code for running LPMVS - Estimates 3D line map for one view
• fuisible/fuisible.cu: File containing cuda code for merging all 3D line per-view maps into one.
• gipuma/scripts/97_frame_00005.sh: Shell script file for executing gipuma for each view and merge the results by running fuisible

Setup

Use cmake to compile fusible and gipuma projects:

1. cd fuisible
2. cmake . -D CMAKE_C_COMPILER=/usr/bin/gcc-7
3. make
4. cd ../gipuma
5. cmake . -D CMAKE_C_COMPILER=/usr/bin/gcc-7
6. make

Data

1. Create a new folder data/my_dataset inside the gipuma folder
2. Add a subfolder images containing the multiview images and the camera parameters following the KRT format
3. [Optional] Add another subfolder for masks contining the hair segmentation masks. masks/cam_img1.png should correspond to images/cam_img1.png, ...
4. [Optional] Add another subfolder for depth contining the initial depth values. depth/cam_img1.dat should correspond to images/cam_img1.png, ...
5. Add another subfolder for orientation_fields contining the generated orientation fields. orientation_fields/cam_img_1_finalOrient.flo should correspond to images/cam_img1.png, ...
6. Add another subfolder for confidence_values contining the generated orientation fields. confidence_values/cam_img_1_finalVariance.flo should correspond to images/cam_img1.png, ...
7. Add a script my_dataset.sh in the /scripts folder to run the Line-based PatchMatch MVS. For an example see: scrips/97_frame_00005.sh

The final file structure should look like:

├── gipuma
|   ├── ...
|   ├── data
|   |   ├── my_dataset
|   |   |   ├── images
|   |   |   |   ├── cam_img_1.png
|   |   |   |   ├── cam_img_2.png
|   |   |   |   ├── ...
|   |   |   |   ├── cam_img_N.png
|   |   |   |   ├── cam_par.txt
|   |   |   ├── depth                             # Optional folder
|   |   |   |   ├── cam_img_1.png
|   |   |   |   ├── cam_img_2.png
|   |   |   |   ├── ...
|   |   |   |   ├── cam_img_N.png
|   |   |   ├── masks                             # Optional folder
|   |   |   |   ├── cam_img_1.png
|   |   |   |   ├── cam_img_2.png
|   |   |   |   ├── ...
|   |   |   |   ├── cam_img_N.png
|   |   |   ├── orientation_fields
|   |   |   |   ├── cam_img_1_finalOrient.flo
|   |   |   |   ├── cam_img_2_finalOrient.flo
|   |   |   |   ├── ...
|   |   |   |   ├── cam_img_N_finalOrient.flo
|   |   |   ├── confidence_values
|   |   |   |   ├── cam_img_1_finalVariance.flo
|   |   |   |   ├── cam_img_2_finalVariance.flo
|   |   |   |   ├── ...
|   |   |   |   ├── cam_img_N_finalVariance.flo
|   ├── ...
|   ├── scripts
|   |   ├── 97_frame_00005.sh
|   |   ├── my_dataset.sh
|   |   ├── ...
|   ├── ...
├── ...

Finally, run the following command: ./my_dataset.sh

The final result will be stored in gipuma/results/{output_dir_basename}/consistencyCheck-.../line_cloud.dat

Running Line Fusion

• CMakeLists.txt: The CMake build configuration file for compiling the program.
• lib/LineFusion.hpp: Header file containing the LineFusion class for performing line fusion on the noisy point cloud.
• pointclouds/: Directory containing example point cloud files.
• src/LineFusion.cpp: Source code file implementing the LineFusion class.
• src/main.cpp: Main source code file implementing the program's main functionality.

Usage

1. Ensure that you have the Point Cloud Library (PCL) and Eigen3 libraries installed on your system.

2. Navigate to the repository's root directory.

3. Create a build directory and compile the program using CMake:

   mkdir build
   cd build
   cmake ..
   make 
   

4. Run the program by providing the input and output filenames as command-line arguments:

   ./line_fusion <lpmvs_result> <output_pointcloud_file>

   Replace <lpmvs_result> with the path to the binary file containing the LPMVS data, and <output_pointcloud_file> with the desired path for the fused point cloud output.

Running Short Hair Generation

• Hair_Generation.cpp: Source code for the hair strands generation.
• all_output_strand.ply: The generation result.

Usage

1. Ensure that you have the Point Cloud Library (PCL), Eigen3 and OpenCV installed on your system.

2. Run the program in Visual Studio 2022