challenge-welding 1.0

Set of tools functions to help challenge Welding users to interacts with challenge datasets

Welding quality detection challenge starter-kit

We recall the website of this challenge: https://confianceai.github.io/Welding-Quality-Detection-Challenge/

For any issues or technical support : please contact [e-mail]

All code present in this repository has been tested with python 3.12.7.

Preparing your environnement

Create and activate your virtual environnement

To create a virtual environnement, you can use many different tools as python virtual environments (venv), conda or uv. Conda and uv has the advantage of making possible to choose the python version you want to use in your virtual env by setting X.Y in the commands below:

If using python integrated venv , to create a new virtual environnement, type in a terminal :

python -m venv path_to_your_env

If using conda :

conda create -n path_to_your_env python=X.Y 

If using uv :

uv venv path_to_your_env --python=X.Y

Activate your virtual env

On Windows power shell

./path_to_your_env/Scripts/activate

On Linux:

source path_to_your_env/bin/activate

Installation of the ChallengeWelding package

To install the package Challenge Welding and its dependencies, from the root directory of this repository type:

pip install .

pip install -r requirements.txt

Dataset Storage organisation

The list of available datasets is present in a yaml file at the following url

https://minio-storage.apps.confianceai-public.irtsysx.fr/challenge-welding/datasets_list.yml

Each dataset has a parquet file that contains the metadata of all samples present in the dataset. This file is accessible for a dataset named : "YOUR_DS_NAME" at :

https://minio-storage.apps.confianceai-public.irtsysx.fr/challenge-welding/datasets/YOUR_DS_NAME/metadata/ds_meta.parquet

Look the Dataset informations for more informations about metadescription contents

Expected AI solution

Look the Solution requirements and evaluation process to get informations about the expected solution and the way it will be evaluated

An exemple of AI Component is provided here, only to show an example of the expected architecture. The model has not been trained to be performant and efficient for this challenge.

• 01-Tutorial.py : This code shows how to use main user functions present in this package. Wihtin, there are examples about how to listing available datasets, explore metadata, and draw basic statistics on contextual variables.
• 02-Create_pytorch_dataloader.py : This code show how to use this package to create a Pytorch dataloader. Run this example will require to install an additionnal dependency torch==2.6.0
• 03-Test_AIComponent.py : This script describe how you can build your own AI component and test it to check if it is compatible with the evaluation pipeline

You have quivalent examples as jupyter-notebooks.

• 01-Tutorial.ipynb :
• 02-Create_pytorch_dataloader.ipynb
• 03-Test_AIComponent.ipynb

Dataset Storage organisation

The list of available datasets is present in a yaml file at the following url

https://minio-storage.apps.confianceai-public.irtsysx.fr/challenge-welding/datasets_list.yml

Each dataset has a parquet file that contains the metadata of all samples present in the dataset. This file is accessible for a dataset named : "YOUR_DS_NAME" at :

https://minio-storage.apps.confianceai-public.irtsysx.fr/challenge-welding/datasets/YOUR_DS_NAME/metadata/ds_meta.parquet

Look the Dataset informations for more informations about metadescription contents