Master Class AI, ESTIA, France, 7-11 April 2026
Computer Vision Applied to a Real Industrial Use Case
This module is reserved for students from partner universities. No extra fees since you're enrolled !
Programme
Use-case description
The use case addressed in this workshop was initially provided in the context of ai4Industry workshop (https://ai4industry.fr/), co-organized by ESTIA. The objective is to extend the work by addressing more advanced technical and methodological aspects.
Context
In the current claims declaration process, several formats for supporting documents are accepted, except for video files. However, policyholder behavior has evolved, with an increasing tendency to submit videos as part of claim reports. To stay aligned with these emerging practices and meet customer expectations, the company must adapt its internal processes accordingly.
Objectives
- Develop automated methods to analyze video content and extract relevant indicators to support claims managers in their assessments.
- Generate representative keyframes from submitted videos for integration into the document management system.
Expected results
A computer vision model capable of processing and analyzing video data to produce the indicators required for effective claim assessment.
Validation criteria
- Achieve a target model accuracy of at least 90%.
- Obtain expert validation of the identified incident typologies.
Data
- Sample videos covering two typical scenarios: road incidents and home incidents.
- A sufficient number of videos with adequate quality and diversity are provided to support experimentation and model development.
Syllabus
Target Audience: Master’s / Engineering students in AI / Data Science.
Prerequisites
Basic knowledge of supervised Machine Learning, Deep Learning,and Python programming (PyTorch/TensorFlow, OpenCV).
Learning Objectives
By the end of the master class, participants will be able to:
• Understand the challenges and industrial applications of computer vision for images and videos.
• Preprocess visual data (images, videos) for analysis.
• Build and train pipelines for image/video classification or detection tasks.
• Experiment with pre-trained models and adapt them to an industrial use case.
• Evaluate model performance and interpret results with respect to business requirements.
Workload distribution
Preparatory work (before the on-site workshop) - 15h
Self-learning with provided resources (readings, tutorials, videos): computer vision fundamentals, CNNs for images/videos. Setting up the Python environment and initial dataset exploration.
On-site sessions – 20h face-to-face + 15h of guided autonomous work
Lectures (face-to-face):
• Applications of computer vision
• Preprocessing and pipelines for images/videos
• Architectures and models: CNNs, YOLO
• Transfer learning strategies
• Evaluation and validation metrics
Hands-on sessions (face-to-face):
• Data preparation and preprocessing
• Building a basic classification/detection model
• Using pre-trained models
• Setting up an end-to-end pipeline and analyzing results
Guided autonomous work:
• Group work on the dataset: experimentation, model adaptation, and validation.
Post-work (after the workshop) – 20h
• Finalization and improvement of the developed pipeline.
• Preparation of deliverables: short written report covering methodology, results, and perspectives.
• Assessment: oral presentation (30 minutes per group)
Expected Outcomes
• A functional computer vision pipeline for images and/or videos.
• A written deliverable documenting methodology, results, and perspectives.
• A short oral presentation of results.
Enrollment
Please contact the head of your current program. For ESTIA students, this program is open to 2nd and 3rd year engineering students, MSc BIHAR students, and doctoral candidates.