brails.utils.model_utils module

This module provides a utility class for computer vision models in BRAILS.

ModelUtils()

Utility class for computer vision models in BRAILS.
class brails.utils.model_utils.ModelUtils

Bases: object

Utility class for computer vision models in BRAILS.

This class provides static methods to ensure necessary model files are available locally, downloading them if needed. Intended for use in applications that rely on pre-trained model weights.

To use ModelUtils, include this import statement in your code:

from brails.utils import ModelUtils
static get_model_path(model_path='', default_filename='', download_url='', model_description='model', overwrite=False)

Check if a model file is available locally, download it if necessary.

Parameters:

  • model_path (str, optional) – Custom path to the model file. If provided, no download occurs.

  • default_filename (str, optional) – Filename to use if downloading the model.

  • download_url (str, optional) – URL to download the model if model does not exist locally.

  • model_description (str, optional) – Human-readable description of the model (default: 'model').

  • overwrite (bool, optional) – If True, re-download and overwrite the model file even if it exists. Defaults to False.

Returns:

Absolute path to the model file.

Return type:

str

Raises:

ValueError – If model_path is not provided and either default_filename or download_url is missing.

Examples

Use a custom model path:

>>> from brails.utils import ModelUtils
>>> path = ModelUtils.get_model_path(
...     model_path='my_models/custom.pth'
... )
Inferences will be performed using the custom model in
my_models/custom.pth

Download a default model if not already available:

>>> path = ModelUtils.get_model_path(
...     default_filename='default_model.pth',
...     download_url=(
...         'https://zenodo.org/record/7271554/files/'
...         'trained_model_rooftype.pth'
...     ),
...     model_description='roof classification model'
... )
Downloading default roof classification model to
tmp/models/default_model.pth...
100%|██████████| 77.9M/77.9M [04:11<00:00, 325kB/s]
Default roof classification model successfully downloaded.

Force overwrite an existing model file:

>>> path = ModelUtils.get_model_path(
...     default_filename='default_model.pth',
...     download_url=(
...         'https://zenodo.org/record/7271554/files/'
...         'trained_model_rooftype.pth'
...     ),
...     model_description='roof classification model',
...     overwrite=True
... )
Re-downloading default roof classification model to
tmp/models/default_model.pth...
100%|██████████| 77.9M/77.9M [04:12<00:00, 324kB/s]
Default roof classification model successfully downloaded.