Mohammad Fahes

CS PhD (Inria)

Research interests: Multimodal Learning, LLMs, Efficient Adaptation, Generalization, Computer Vision, Deep Learning

More broadly, I'm interested in machine learning, mathematics, epistemology, and science in general.

News

09/2026
I defended my PhD, entitled Efficient Adaptation of Vision-Language Models.
04/2026
11/2025
ProLIP is accepted at WACV 2026.
10/2025
Outstanding Reviewer award at ICCV 2025.
08/2025
I started a 6-month internship at Google DeepMind, working with Cordelia Schmid and Anurag Arnab.
06/2025
FLOSS is accepted at ICCV 2025.
05/2025
Outstanding Reviewer award at CVPR 2025.
09/2024
Outstanding Reviewer award at ECCV 2024.
02/2024
FAMix is accepted at CVPR 2024.
07/2023
PØDA is accepted at ICCV 2023.
10/2022
I started my PhD at Inria.
01/2022
LPALM is accepted at ICLR 2022.
04/2021
I started my final-year internship at Télécom Paris.

Publications

Domain Adaptation with a Single Vision-Language Embedding

Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc, Patrick Pérez, Raoul de Charette

IJCV 2026

A unified framework for domain adaptation that replaces all target data with a single vision-language latent vector.

CLIP's Visual Embedding Projector is a Few-shot Cornucopia

Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc, Patrick Pérez, Raoul de Charette

WACV 2026

Few-shot and test-time adaptation of VLMs need not be complicated: regularized fine-tuning of their projection matrix suffices.

FLOSS: Free Lunch in Open-vocabulary Semantic Segmentation

Yasser Benigmim, Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc, Raoul de Charette

ICCV 2025

Improving open-vocabulary semantic segmentation with unsupervised prompt selection.

A Simple Recipe for Language-guided Domain Generalized Segmentation

Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc, Patrick Pérez, Raoul de Charette

CVPR 2024

Multi-source domain generalization can be replaced by a single source domain combined with random language prompts.

PØDA: Prompt-driven Zero-shot Domain Adaptation

Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc, Patrick Pérez, Raoul de Charette

ICCV 2023

A new framework for domain adaptation driven by a single natural-language prompt.

Unrolling PALM for Sparse Semi-blind Source Separation

Mohammad Fahes, Christophe Kervazo, Jérôme Bobin, Florence Tupin

ICLR 2022

An unrolled version of the PALM algorithm that reduces its sensitivity to initialization and improves source separation quality.

Academic Services

Conference Reviewer
CVPR 2024–2025, ECCV 2024, ICLR 2025, ICCV 2025, NeurIPS 2025–2026, ICLR 2027.
Journal Reviewer
IEEE Transactions on Multimedia.