Sylvain Paris, Adobe
Will Chang, UCSD
Wojciech Jarosz, UCSD
Oleg Kozhushnyan, MIT CSAIL
Wojciech Matusik, Adobe
Matthias Zwicker, UCSD
Frédo Durand, MIT CSAIL

Geometric Data

• Straight Hair

Reconstructed Hair Strands Head Model

• Tangled Hair

Reconstructed Hair Strands Head Model

• Puffy Hair

Reconstructed Hair Strands Head Model

• Wavy Hair

Reconstructed Hair Strands Head Model

Strand Data Format

The hair strand data is made of integers and floating point values. Both types of data are coded on 4 bytes. A hairstyle is made of strands. A strand is made of vertices. A vertex is made of (x,y,z) coordinates. The actual format is:

    [INT: number of strands]

      [INT: number of vertices in strand 1]

        [FLOAT: x1] [FLOAT: y1] [FLOAT: z1]
        [FLOAT: x2] [FLOAT: y2] [FLOAT: z2]
        ...

      [INT: number of vertices in strand 2]

        [FLOAT: x1] [FLOAT: y1] [FLOAT: z1]
        [FLOAT: x2] [FLOAT: y2] [FLOAT: z2]
        ...

      ...

The following C++ code reads this file format. We assume that there is a type implementing a 3D vector (in our case Geometry::Vec3f).

std::deque<Geometry::Vec3f> vertices;        /**< Hair vertices */
std::deque<std::deque<int> > lines;          /**< Hair strands */

void load_hairstyle( const char *hairFile )
{
    ifstream in;
    in.open(hairFile, ios_base::binary);

    unsigned int nStrands;
    int currentVertex = 0;

    in.read((char*)&nStrands, sizeof(unsigned int));
    cout << "Reading " << nStrands << " strands." << endl;

    for (unsigned int strand = 0; strand < nStrands; strand++) {
        unsigned int numVertices;
        in.read((char*)&numVertices, sizeof(unsigned int));

        // For each strand, first read all of the vertices
        for (unsigned int i = 0; i < numVertices; i++) {
            float x, y, z;
            in.read((char*)&x, sizeof(float));
            in.read((char*)&y, sizeof(float));
            in.read((char*)&z, sizeof(float));

            Geometry::Vec3f vertex(x, y, z);
            vertices.push_back(vertex);
        }

        // Then store a line containing all of those vertices
        deque<int> line;
        for (unsigned int vertex = 0; vertex < numVertices; vertex++) {
            line.push_back(currentVertex++);
        }
        lines.push_back(line);
    }

    in.close();
}

Photometric Data

Each dataset contains two capture sequences, with the following images:

1. 64 Black images: calibration images taken with no lighting.
   Filename format, where ## represents the camera number (0--15):

Capture Sequence 0 (32 images): black_0_0_##.png, black_0_1_##.png
Capture Sequence 1 (32 images): black_1_0_##.png, black_1_1_##.png

2. 4800 Hair images: images from each camera taken with each light turned on in succession.
   Filename format, where ### represents the light number (0--149) and ## represents the camera number (0--15):

   Capture Sequence 0 (2400 images): refl0_###_##.png
   Capture Sequence 1 (2400 images): refl1_###_##.png

The files are split into 5 parts each. In order to combine them into a single tar archive, use the command:

$ cat refl-name_aa refl-name_ab refl-name_ac refl-name_ad refl-name_ae > refl-name.tar

where "name" is the name of the dataset.
Each image was taken at 1300x1030 resolution.

Acknowledgement