Computer Vision Engineer @ Quividi - 01/2024-07/2024 - Paris, France

Context

In Quividi’s R&D department, I work on AI models for gender classification.

Missions

• Evaluate the effectiveness of CNN-based models and compare them to two deployed models: an outdated obsolete model based on a traditional machine-learning method (fast but less accurate) and a state-of-the-art model (accurate but slower).
• Comparison of state-of-the-art classification methods.
• Training and evaluation of gender classification models.
• Integration of the prototype into the main software.

Results

• Comparison of state-of-the-art classification methods and recommendation of architectures for edge devices.
• Creation of an automatic system for data labeling.
• Created an AI factory framework for training/fine-tuning models.
• The proposed model offered a better compromise between speed and accuracy than the two other models.
• Integration of the model into the software

Technical Environment

Python, Jupyter, PyTorch, OpenCV, dlib, numpy, matplotlib, TensorBoard, Git, C++, OpenVino, Linux