Internship - VLM calibration and finetuning
Since our founding, IDEMIA has been on a mission to unlock the world and make it safer through our cutting-edge identity technologies. Our technology leadership makes us the partner of choice for hundreds of governments and thousands of enterprises in over 180 countries, including some of the biggest and most influential brands in the world. In applying our unique expertise in biometrics and cryptography, we enable our clients to unlock simpler and safer ways to pay, connect, access, identify, travel and protect public places – at scale and in total security.
Our teams work from 5 continents and speak 100+ different languages. We strongly believe that our diversity is a key driver of innovation and performance.
Purpose
IDEMIA has a long history in assessing and mitigating AI risks. Leveraging this expertise, we deliver solutions that provide measurable confidence in algorithm outputs, essential for critical deployments, while upholding the highest ethical standards, including algorithmic fairness.
Beyond traditional biometric retrieval tools, the recent breakthroughs in multimodal models such as CLIP and its successors have opened new possibilities for semantic search. Although CLIP delivers impressive retrieval performance, its score distribution is not well‑understood, which hampers accurate thresholding and reduces effectiveness in large‑scale Re‑Identification (ReID) scenarios. Robust CLIP‑based retrieval is also a critical component of Retrieval‑Augmented Generation (RAG) frameworks, enabling Vision‑Language Models and visual‑understanding systems to scale without succumbing to hallucinations.
This research-focused internship will start with a literature review and replication of state-of-the-art results. The specific research directions will be further refined with input from the intern, based on emerging findings.
About IDEMIA
IDEMIA is the global leader in trusted identities, with the ambition to empower citizens and consumers to interact, pay, connect, travel, and vote in a connected environment. IDEMIA employs over 15,000 people from more than 80 nationalities and serves clients in 180 countries. The Research Unit at IDEMIA is responsible for developing all new technologies that will be included in all of IDEMIA's product lines. Composed of approximately 100 researchers and developers, the unit works on multiple application areas: facial recognition, fingerprint recognition, iris recognition, fraud detection, secure acquisition systems, video analysis, and document recognition and authentication. The URT develops algorithms suited for multiple use cases:
- Identity in the context of civil applications (elections, biometric documents, etc.)
- Border control (fraud detection on documents or biometrics, facial recognition with comparison to passports and reality, etc.)
- Law enforcement/Justice (identifying criminals, object identification in videos, etc.)
- Access control (entry into buildings or any secure area)
- Digital identity on smartphones (unlocking or payment using a smartphone with biometrics, opening a bank account remotely, etc.)
- Embedded terminals or smart cards (payment with a biometric credit card, etc.)
- Road safety (smart radars, etc.)
Key Missions
- Litterature Review on CLIP-Base models (Siglip, AlphaClip, ...) and the dynamic of their distribution scores
- Implementing Alignement method and finetuning for improving thresholded accuracy of CLIP-base models
- Develops test strategies and methodologies to evaluate new or competing algorithms
- Stays up-to-date on latest state of the art research and algorithm development methodologies
- Collaborating with researchers to refine methodologies and contribute to the overall project goals
- Documenting findings and presenting results to stakeholders
Profile & Other Information
Education: Engineering student in the final year of the engineering cycle, or a Master's student (M2) specializing in computer vision, image processing, or Deep Learning.
Required Technical Skills:
- Strong background in mathematics, data analysis, Python, and PyTorch
- Ability to handle large-scale data and work in a research environment (exploring unfamiliar domains)
- Knowledge in biometrics is a plus
Soft Skills & Languages:
- Curious
- Proactive and autonomous
- Results- and solutions-oriented
- Clear and convincing communication
- Fluent spoken and written English
By choosing to work at IDEMIA, you will join a unique tech company, offering a wide range of growth opportunities. You will contribute to a safer world, collaborating with an international and global community. We value the diversity of our teams and welcome people from all walks of life, regardless of how they look, where they come from, who they love, or what they think.
We deliver cutting edge, future proof innovation that reach the highest technological standards and we’re transforming, fast, to stay a leader in a world that’s changing fast, too.
At IDEMIA, people can develop their expertise and feel a sense of ownership and empowerment, in a global environment, as part of a company with the ambition and the ability to change the world.
Visit our website to know more about the leader in Identity Technologies