README.md

CableInspect-AD Dataset

Download the dataset and extract files:

wget https://hydroquebec.com/data/documents-donnees/donnees-ouvertes/zip/CableInspect-AD.zip
unzip CableInspect-AD.zip
cd CableInspect-AD
tar -xzvf cable_1.tar.gz && rm cable_1.tar.gz
tar -xzvf cable_2.tar.gz && rm cable_2.tar.gz
tar -xzvf cable_3.tar.gz && rm cable_3.tar.gz

The CableInspect-AD dataset folder is structured as follows:

CableInspect-AD
    ├── cable_1.json
    ├── cable_1_seg.json
    ├── cable_2.json
    ├── cable_2_seg.json
    ├── cable_3.json
    ├── cable_3_seg.json
    ├── Cable_1
    ├── Cable_2
    ├── Cable_3
    ├── licence.txt
    └── readme.txt

Dataset Pre-processing

To pre-process the dataset, create the following conda environment:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh -b -p $HOME/miniconda3
conda update conda
conda create -n cableinspect_ad python=3.10
conda activate cableinspect_ad
pip install -r dataset/requirements.txt

In the following instructions, we assume the dataset is stored in the $HOME directory.

To prepare the dataset, run the following command from the cableinspect-ad-code root directory:

bash dataset/tools/dataset.sh

This will create in the CableInspect-AD directory with the following structure:

CableInspect-AD
    └── Cable_1
        └──images
            └── 01
                └──*.png
            └── 02
                └──*.png
            └── 03
                └──*.png
        └──masks
            └── 01
                └──*.png
        └── Cable_2
            └── ...
        └── Cable_3
            └── ...
        ├── labels.csv
        ├── k_fold_labels
        ├── licence.txt
        └── readme.txt

Below is an example row of the labels.csv file:

image_path	cable_id	side_id	pass_id	frame_id	bbox_area	bbox_x	bbox_y	bbox_width	bbox_height	bbox_rotation	anomaly_type_id	anomaly_type	anomaly_grade	anomaly_id	label_index	mask_path
<image_path>	C01	A	1	0	661.5496	240.32	691.39	26.98	24.52	0.0	7	Deposit	light	003_01	1	<mask_path>

The k_fold_labels folder has the following structure:

k_fold_labels
└── C01
    └── 2
        ├── label_cable-C01_num_k_shot-2_anomaly_id-0.csv
        ├── label_cable-C01_num_k_shot-2_anomaly_id-1.csv
        └── ...
    └── 3
        ├── label_cable-C01_num_k_shot-3_anomaly_id-0.csv
        ├── label_cable-C01_num_k_shot-3_anomaly_id-1.csv
        └── ...
    └── ...

└── C02
    └── 2
        ├── label_cable-C02_num_k_shot-2_anomaly_id-0.csv
        ├── label_cable-C02_num_k_shot-2_anomaly_id-1.csv
        └── ...
    └── 3
        ├── label_cable-C02_num_k_shot-3_anomaly_id-0.csv
        ├── label_cable-C02_num_k_shot-3_anomaly_id-1.csv
        └── ...
    └── ...

└── C03
    └── 2
        ├── label_cable-C03_num_k_shot-2_anomaly_id-0.csv
        ├── label_cable-C03_num_k_shot-2_anomaly_id-1.csv
        └── ...
    └── 3
        ├── label_cable-C03_num_k_shot-3_anomaly_id-0.csv
        ├── label_cable-C03_num_k_shot-3_anomaly_id-1.csv
        └── ...
    └── ...

Below is an example row of the label_cable-{cable_id}_num_k_shot-{num_k_shot}_anomaly_id-{anomaly_id}.csv files:

cable_id	side_id	pass_id	image_path	mask_path	label_index	split
C01	A	1	<image_path>	<mask_path>	1	test

Cropped Version of the Dataset

To prepare the cropped version of the dataset, run the following command from the cableinspect-ad-code root directory:

bash dataset/tools/cropped_dataset.sh

This will create a CableInspect-AD-cropped folder with the same structure as the CableInspect-AD folder after preprocessing.