Implementation of SIWNet from the paper "Enhanced Winter Road Surface Condition Monitoring with Computer Vision".
Run the script with:
pip3 install -r requirements
To train and validate, run:
python3 train.py -p <path-to-params> -tr <path-to-train-csv> -v <path-to-val-csv> -s <path-to-save-directory> -n <name-of-run>
To train and test, run:
python3 train.py -p <path-to-params> -tr <path-to-train-csv> -v <path-to-val-csv> -te <path-to-test-csv> -s <path-to-save-directory> -n <name-of-run>
For inference, run:
python3 inference.py -wb <path-to-basenet-weights> -wp <path-to-pihead-weights> -i <path-to-image>
For testing, the model is trained with both training and validation data.
Example of the parameter format is provided in params/example_params.json.
The training/validation/testing data should be provided as a .csv-files, which are formatted as
<path-to-image>, <grip-factor-value>