Martin Nicolas Everaert Martin Nicolas Everaert Final-year PhD candidate · EPFL IVRL

Martin Nicolas Everaert

Research: Computer Vision and Generative AI, focus on Text-to-Image Diffusion Models

Final-year PhD candidate in the Image and Visual Representation Lab (IVRL) at EPFL.

News

Jun 2026

I volunteered to serve as reviewer for NeurIPS 2026, currently working through my assigned reviews.

Jun 2026

I presented Visual Grounding for Object Questions at CVPR@Paris 2026, and prepared the poster and video for the main CVPR 2026 in Denver, where my co-authors are presenting it.

May 2026

🎓 Passed my private PhD defense.

Feb 2026

My internship paper, Visual Grounding for Object Questions, was accepted to the CVPR 2026 main track.

Dec 2025

I wrapped up a six-month Applied Science internship at Amazon in Sunnyvale and returned to Lausanne.

About me

I am currently a final-year PhD candidate in the Image and Visual Representation Lab (IVRL) at EPFL. I am grateful to be supervised by Prof. Sabine Süsstrunk and Dr. Radhakrishna Achanta.

Education

20262021Image and Visual Representation Lab (IVRL), EPFL (Lausanne, Switzerland)
PhD in Computer Science
20212019EPFL (Lausanne, Switzerland)
Master's degree in Computer Science
20212017CentraleSupélec (Paris-Saclay, France)
Diplôme d'Ingénieur (combined BS/MS in Engineering)
20182017Paris-Sud (Paris XI) University (France)
Bachelor's degree in Fundamental Physics
20172015Lycée Saint-Louis (Paris, France)
CPGE (pre-college school), MPSI / MP*

Experience

2025Amazon (Sunnyvale, CA, USA)
6-month Applied Science internship at Amazon.com Services LLC
2021Thales LAS (Élancourt, France)
6-month internship / Master project at Thales Land and Air Systems
2019CEA (Paris-Saclay, France)
3-month engineering internship at CEA (Saclay Nuclear Research Center)
2018TCL (Chengdu, China)
1-month internship at TCL
More about me

I am interested in Computational Aesthetics, overlapping with Computation and Language, Computer Vision, Deep Learning, Generative Models, etc. My research specifically focuses on Text-to-Image Diffusion Models.

Prior to pursuing my PhD, I completed a Master's in Computer Science at EPFL in 2021. I also hold a Diplôme d'Ingénieur (engineering BS/MS degree) from the Supélec curriculum at CentraleSupélec, graduating concurrently in 2021, and a BS degree in Fundamental Physics from Université Paris-Sud in 2018. Before that, I attended Lycée Saint Louis, where I pursued French preparatory classes (CPGE), with a focus on Math, Physics, Engineering and Computer Science (MPSI/MP*).

Throughout my studies, I interned at Amazon (6-month Applied Science Internship in Sunnyvale, USA, 2025), Thales Group (6-month Master's thesis project in Elancourt, France, 2021), CEA (3-month Engineering internship in Paris-Saclay, France, 2019), and TCL Corporation (1-month Work/Labor internship in Chengdu, China, 2018).

Publications and Patents

Visual Grounding for Object Questions
M. N. Everaert, X. Liu, H. Takeda, R. Bala, V. Yadav, V. Narayanan
CVPR 2026
Visual grounding has focused on directly visible elements (e.g., 'what is the title of this book?'). We introduce a new task: grounding the visual context or evidence that helps answer open-ended object questions.
Covariance Mismatch in Diffusion Models
M. N. Everaert, S. Süsstrunk, R. Achanta
Infoscience preprint 20.500.14299/242173, 2024
We identify a covariance mismatch between the data and the noise distributions in diffusion models. Due to this mismatch, noise affects some components of the data distribution much more than others during training. By realigning these covariances, we improve the model's flexibility and enable better generation.
Exploiting the Signal-Leak Bias in Diffusion Models
M. N. Everaert, A. Fitsios, M. Bocchio, S. Arpa, S. Süsstrunk, R. Achanta
WACV 2024 (accepted in round 1)
Diffusion models are trained on noisy images, but inference starts with pure noise. This negatively affects generating images of a specific style, because noisy images and pure noise have different average colors. We show how to fix this bias without additional training, and how to exploit it to generate better images.
VETIM
M. N. Everaert, M. Bocchio, S. Arpa, S. Süsstrunk, R. Achanta
BMVC 2023
We learn new tokens in the text-encoder of the diffusion model without having to learn new visual features from exemplar images.
Diffusion in Style
Diffusion in Style [ICCV 2023 + Patent]
M. N. Everaert, M. Bocchio, S. Arpa, S. Süsstrunk, R. Achanta
Fine-tuning a diffusion model on a specific style is more efficient if we also adapt the noise distribution to the style.
Estimating Image Depth in the Comics Domain
D. Bhattacharjee, M. N. Everaert, M. Salzmann, S. Süsstrunk
WACV 2022
Our depth estimation model for comics images uses an image-to-image translation GAN and a context-aware depth model trained on natural images, which improves depth accuracy.
See more publications
Aesthetics-Oriented Video Generation and Editing
M. N. Everaert
EDIC research proposal, 2022
Scene relighting with illumination estimation
A. P. Dherse*, M. N. Everaert*, J. J. Gwizdała*
arXiv preprint arXiv:2006.02333, 2020
* equal contribution

Academic activities

Conference service
Social Media Chair of ICCP 2024.
Reviewer and/or emergency reviewer for ICCV 2023, BMVC 2023, ECCV 2024, NeurIPS 2024, CVPR 2025, TMLR, CVPR 2026, and NeurIPS 2026.
Top reviewer at NeurIPS 2024 and outstanding reviewer at CVPR 2025.
more details
Social media Chair of ICCP 2024 · ICCP website · X · YouTube · Facebook
Reviewer for NeurIPS 2026 · (2 ongoing reviews)
Reviewer for CVPR 2026 · (1 review)
Reviewer for TMLR · (2 reviews)
Reviewer for CVPR 2025 · (4 reviews, ✨ recognized as an outstanding reviewer ✨)
Reviewer for NeurIPS 2024 · (6 reviews, ✨ recognized as a top reviewer ✨)
Reviewer + emergency reviewer for ECCV 2024 · (2 reviews, 1 emergency review)
Emergency reviewer for BMVC 2023 · (9 emergency reviews)
Reviewer for ICCV 2023 · (1 review, 1 emergency review)
You can find my name on the CVPR 2025 outstanding reviewers list, the NeurIPS 2024 top reviewers list, the ECCV 2024 emergency reviewers list, the BMVC 2023 emergency reviewers list, and the ICCV 2023 reviewers list.
Talks
Live talks: candidacy exam (2022), AI Tinkerers meetup (2024), theater performance (2025)
more details
🎭 Art/science collaboration for the theater performance Écrire avec des algorithmes, with theater director Marielle Pinsard and collective Quinté+, staged at La Grange (UNIL) · March 2025 · news · event page · theater program
AI Tinkerers Lausanne Inaugural Meetup · June 2024 · news · event page · LinkedIn post
Public presentation of candidacy exam · August 2022 · news · event page · writeup
Prerecorded presentations: CVPR 2026, ICCV 2023, BMVC 2023, WACV 2024, ICCP 2024
more details
Visual Grounding for Object Questions · CVPR 2026 · video · Paper video
Controlling Style in Diffusion Models through the Noise · ICCP 2024 · video · (1-min research video)
Exploiting the Signal-Leak Bias in Diffusion Models · WACV 2024 · video · Paper video
VETIM: Expanding the Vocabulary of Text-to-Image Models only with Text · BMVC 2023 · video · Paper video
Diffusion in Style · ICCV 2023 · video · Paper video
Posters: CVPR 2026, CVPR@Paris 2026, ICCV 2023, BMVC 2023, WACV 2024, EDIC Open House 2024, ICVSS 2024, ICCP 2024
more details
Visual Grounding for Object Questions, at CVPR 2026 · poster · Paper poster
Visual Grounding for Object Questions, at CVPR@Paris 2026 · accepted posters list · Paper poster
Controlling Style in Diffusion Models through the Noise, at ICCP 2024 · poster · Research poster, ✨ ICCP 2024 Spotlight poster
Controlling Style in Diffusion Models through Noise, at ICVSS 2024 · poster · Research poster
Image Generation with Diffusion Models, at EDIC Open House 2024 · poster · Research poster
Exploiting the Signal-Leak Bias in Diffusion Models, presented virtually at WACV 2024 · poster · Paper poster
VETIM: Expanding the Vocabulary of Text-to-Image Models only with Text, at BMVC 2023 · poster · Paper poster
Diffusion in Style, at ICCV 2023 · poster · Paper poster
Awards
Teaching
CS-413: Computational Photography (MS level), taught by Prof. Sabine Süsstrunk · Spring 2022, Spring 2023 (TA Award!), Spring 2024, Spring 2025
CS-119g: Information, Computation, Communication (BS level, Life Science Engineering section) · Fall 2024
CS-328: Numerical Methods for Visual Computing and Machine Learning (BS level), taught by Prof. Wenzel Jakob · Fall 2021, Fall 2023
Supervision
Semester & research projects: project ideas designed by me (co-designed when co-supervised), carried out by BS or MS students.
more details
1 MS student. Development in Diffusion Models: Implementing an "Initial Noise Loader" in the Diffusers Library, Master research project · Spring 2025 · project description · HuggingFace page
1 MS student. Studying the manifold of natural images through the lens of Diffusion Models, Master research project · Fall 2024 · project description
1 MS student. Creating 3D bricks designs with Diffusion Models, Master research project · Fall 2024 · project description · HuggingFace page · co-supervised with Eric Bezzam
1 MS student. Using Large Language Models (LLMs) as Diffusion Models for Image Generation, Master research project · Spring 2024 · project description · co-supervised with Dongqing Wang
1 BS student. Exploring the Benefits of 2D Gaussian Splatting in Image Representation and Compression, Bachelor research project · Spring 2024 · project description · project report
1 BS student. Feed-Forward Guidance for Text-to-Image Diffusion Models, Bachelor research project · Fall 2023 · project description
1 MS student. Generating stories from keywords, Master semester project · Spring 2023 · project description
1 BS student. Coreference resolution for story visualization, Bachelor semester project · Spring 2023 · project description
1 MS student. Generate images from texts, Master semester project · Fall 2022 · project description
1 BS student. Is this video an ad?, Bachelor semester project · Fall 2022 · project description
1 team of 3 MS students. Predicting if two frames are part of the same video, ML4Science project · Fall 2022 · project report · project code
Computational Photography projects: project ideas designed by me, carried out by (mostly) MS students.
more details
1 team of 2 MS & 1 BS students. Automatic and aesthetic image cropping · Spring 2025 · project description
1 team of 3 MS students. Double exposure effect with diffusion model · Spring 2025 · project description
1 team of 3 MS students + 1 team of 2 MS & 1 PhD students. 360°-photography: Stitching dual fish-eye images · Spring 2024 · project description
1 team of 3 MS students. Computational photography with smartphone lenses · Spring 2024 · project description
1 team of 3 MS students. Automatic and Personalized Tunnel Book Generation from Photographs · Spring 2024 · project description
1 team of 1 MS student. 2D Gaussian Splatting for Image Representation and Compression & Text to 2D Gaussian Splatting · Spring 2024 · project description
1 team of 3 MS students. Aesthetics of sets of images · Spring 2023 · project description
1 team of 3 MS students + 1 team of 2 MS students. TimeWarp: How would this scene look in 100 years? · Spring 2023 · project description
2 teams of 3 MS students. Text to Photomosaic · Spring 2023 · project description
1 team of 3 MS students. From RGB to NIR with Style Transfer · Spring 2023 · project description
1 team of 3 MS students. Improving an open-source microscope by adding multispectral imagery · Spring 2022 · project description
2 teams of 3 MS students. Low-level image transformations for image aesthetics · Spring 2022 · project description