The Second International Scan-to-BIM Competition

Deadlines:

Apr. 1st: training/validation data available for 2D, training data available for 3D

May 1st: validation data available for 3D

June 3rd: FINAL SUBMISSION

June 12 11:59 pm PDT: FINAL SUBMISSION (Extended)

Format of your submission:

Please compress all your answer files into one zip file, and use the zip file as your submission.

Classes columns, doors, and walls are accepted and will be parsed.

Files format regex:

(?P<model>.*)_(?P<floor>.*)_(?P<classname>columns|doors|walls).json.

Examples:

MedOffice_F1_doors.json, MedOffice_F2_columns.json, Minas_Tirith_Outerwall_walls.json.

All the files must be placed together in one directory or .zip archive.

Notes:

• Each JSON file is a List of structures (Dict).
• All the measures, locations, and points must be measured in meters! Rotation parameter measured in degrees.

Schemas

Note: The rotation parameter is optional, it will default to 0 if you don’t specify it 

""" Columns.
    Have only one location point and 3D measures:
        Width - X,
        Depth - Y,
        Height - Z.
    Rotation parameter is used to rotation the structure around Z-axis."""
{
    "width": 0.3047,    # Meters
    "depth": 0.2031,
    "height": 2.7432,
    "loc": [
        5.8154,         # X-axis
        9.8700,         # Y-axis
        0.0             # Z-axis
    ],
    "rotation": 0.0     # Degrees
}

""" Doors.
    The same schema as the one above."""
{
    "width": 1.0668,
    "depth": 0.1778,
    "height": 2.2097,
    "loc": [
        4.2008,
        17.7022,
        -0.911
    ],
    "rotation": 359.9999,
}

""" Walls.
    Consists of the two points and width, height dimensions.

        •-------------•
       /|            /|         z    y
      / |           / |         |   /
     /  |          /  |  height |  / width (around ep-st vector)
    •-------------• --•         | /
    |  /          |  /          |/
    | st----------|-ep          •----------x
    |/            |/               length (ep-st)
    •-------------•

    Height measure for both points is assumed to be the same.
"""
{
    "start_pt": [
        35.9333,
        22.5964,
        -0.9111
    ],
    "end_pt": [
        35.9330,
        17.7753,
        -0.9111
    ],
    "width": 0.2032,    # X or Y-axis respectively.
    "height": 3.9529,   # Z-axis
}

Run in Docker

You may want to configure some parameters in src/config.py before build.

# Clone the repository.
git clone https://github.com/arkhodakov/cvpr-2022-matching
cd cvpr-2022-matching

# Build Docker image.
docker build -t cvpr-2022-matching .

# Run image and evaluate.
# NOTE: Put all your JSON data in ./data/.
docker run \
    -v {full/path/to/ground-truth/directory/}:/data/ \
    -v {full/path/to/users-models/directory/}:/predicted/ \
    -v {full/path/to/output/directory/}:/code/output \
    -it cvpr-2022-matching /bin/bash

python main.py ../data/{reference_model}.json ../predicted/{user_model}.json --output/match.json

# Example:
python main.py ../data/OfficeLab01_Allfloors_columns.json ../predicted/OfficeLab01_Allfloors_columns.json --output output/match.json

Note: you need to pass a directory path or a file path to the script.

• When the path is a (like ./data) directory the script searches for all models inside one level deep.
• When the path is a file (like *_columns.json) the script uses regular expressions to extract the model name of the file. Then it searches for all the data for that model inside the directory. It means that if you pass ../data/OfficeLab01_Allfloors_columns.json file to the script, it will search for all the files with OfficeLab01 model name around.

Links:

Workshop

2D challenge

 3D Challenge

Point Cloud Files for the 3D Training Set

Ground Truth for the 3D Training set

Point Cloud Files for the 3D Testing Set

3D Challenge Evaluation Toolkit