Skip to content

[CV4AEC Workshop CVPR 2023] 3D building models matching & evaluation code

Notifications You must be signed in to change notification settings

cv4aec/3d-matching-eval

Repository files navigation

3D Matching

Data Preparation

Classes columns, doors, 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 files is a List of structures (Dict).
  • All the measures, locations and points must be measured in meters! Rotation parameter measures in degrees.
  • All parameters are optional with 0 default value. You may not specify them

Schemas

""" 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,    # Optional,
    "depth": 0.2031,    # default
    "height": 2.7432,   # is 0
    "loc": [
        5.8154,         # X-axis
        9.8700,         # Y-axis
        0.0             # Z-axis
    ],
    "rotation": 0.0     # Degrees (Optional: default is 0)
}

""" Doors.
    The same schema as the one above.
    Translation and rotation parameters are optional."""
{
    "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.
    Translation parameters are optional.
"""
{
    "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 buid.

# Clone the repository.
git clone [email protected]:cv4aec/3d-matching-eval.git
cd 3d-matching-eval

# Build Docker image.
docker build -t 3d-matching-eval .

# 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 3d-matching-eval /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.

About

[CV4AEC Workshop CVPR 2023] 3D building models matching & evaluation code

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published