Hatef Otroshi

Researcher in Computer Vision and Machine Learning | IEEE Senior Member | Postdoc at @Idiap_ch | PhD from @epfl_en

[Hatef Otroshi @HOtroshi]

@NeurIPSConf Last year, only 40 papers out of near 700 papers were accepted (<6%). Is there any change in the acceptance policy (ie, limited number of papers or very small acceptance rate, like 2025)?

[Hatef Otroshi @HOtroshi]

📄 Pre-print (arxiv): https://t.co/GYY5fVCxLe
🌐 Project page: https://t.co/uCkgelUVKo
💻 Code: check our project page!

This is a joint work with Unsal Ozturk and Sébastien Marcel. Congratulations to the team! 🚀

🧵(3/3)

#bias #fairness #LLM #VLM #MLLM

[Hatef Otroshi @HOtroshi]

Multimodal LLMs are rapidly being adopted for vision-language tasks applications. But how fair are they across demographic groups?🤔

🔍In our recent work, accepted in #CVPR 2026 workshops, we evaluate gender and ethnicity bias in 9 different MLLMs for face verification.

🧵(1/3)

[Hatef Otroshi @HOtroshi]

📊Key findings:
✅ Bias persists in MLLMs: Performance varies significantly across demographic groups
✅ Gender and ethnicity disparities can be observed even in advanced models
✅ There is a gap between generative capability and trustworthy decision-making
🧵(2/3)#bias #fairness

[Hatef Otroshi @HOtroshi]

🚨 Deadline Extension! 🚨

The submission deadline for @CVPR Workshop on Foundation and Generative Models in Biometrics is extended to March 10.

🌐 Visit the workshop page for more details and call for papers: https://t.co/QLapcQaGnz

#CVPR #CVPR26 #CVPR2026

[Hatef Otroshi @HOtroshi]

📢Thrilled to announce that we are organising the 2nd Workshop on Foundation and Generative Models in Biometrics at #CVPR 2026 !

📍Denver, USA - @CVPR
🗓️June 3-4, 2026

🔍More details and call for papers: https://t.co/QLapcQaGnz

[IEEE Signal Processing Society @IEEEsps]

✨Paper spotlight!

The paper provides an in-depth analysis of state-of-the-art methodologies regarding foundation multi-modal models, their advancements, and their applicability to biometrics tasks.

🔗Read at https://t.co/1ztPbWXkjQ.

[Hatef Otroshi @HOtroshi]

@giffmana @paperswithcode Yes indeed! I noticed it a week ago:

[Hatef Otroshi @HOtroshi]

@paperswithcode Not only badges are broken again, but the website (https://t.co/1cQt06j8gd) is now down!

@paperswithcode #paperswithcode

[Hatef Otroshi @HOtroshi]

@paperswithcode Not only badges are broken again, but the website (https://t.co/1cQt06j8gd) is now down!

@paperswithcode #paperswithcode

[Hatef Otroshi @HOtroshi]

Is there any issue with paperswithcode? All badges seem broken!
@paperswithcode #paperswithcode

[Hatef Otroshi @HOtroshi]

@paperswithcode The badges are now back live! Thanks for the quick fix!

[Hatef Otroshi @HOtroshi]

@_onionesque Not yet! @NeurIPSConf #NeurIPS

[Hatef Otroshi @HOtroshi]

🏆Our experiments demonstrate that FaceLLM achieves the state-of-the-art performance of MLLMs on various face-centric tasks🎭.

[Hatef Otroshi @HOtroshi]

Excited to share our new paper, "FaceLLM: A Multimodal Large Language Model for Face Understanding" ! ✨

📄 Pre-print(arxiv): https://t.co/7Et8FNt4ad
🌐 project page: https://t.co/JwzsWLK5PB

[Hatef Otroshi @HOtroshi]

We introduce FaceLLM, a MLLM trained specifically for facial image understanding. To construct the training data, we propose a novel weakly supervised pipeline that generates high-quality question-answer pairs based on images from the FairFace dataset.

[Hatef Otroshi @HOtroshi]

🚨 Deadline Extension! 🚨

The submission deadline for #ICCV2025 Workshop on Foundation and Generative Models in Biometrics is extended to July 1. @ICCVConference #ICCV

🔍Visit the workshop page for more details and call for papers: https://t.co/GThMMCUktq

[Hatef Otroshi @HOtroshi]

📢We are thrilled to announce that we are organising the Workshop on Foundation and Generative Models in Biometrics at #ICCV 2025 !

📍Honolulu, Hawaii - @ICCVConference
🗓October 19-20, 2025
#ICCV2025

🔍More details and call for papers: https://t.co/GThMMCUktq

[Hatef Otroshi @HOtroshi]

🧵(4/4) We also show that our approach can be used to mitigate leakage from the generator's training set and explore the ability of generative models to generate data beyond it. #ICML #ICML2025

[Hatef Otroshi @HOtroshi]

Excited to share that our paper is accepted at #ICML 2025 !

We introduce a novel physics-inspired approach based on Brownian motion to generate synthetic face recognition datasets.

📄 Pre-print(arxiv): https://t.co/An5DsZbgrW
🌐 project page: https://t.co/tkvChOo0lB

🧵(1/4)

[Hatef Otroshi @HOtroshi]

🧵(3/4) We generate several synthetic face datasets and evaluate them by training face recognition models and benchmarking trained models on real face recognition benchmark datasets. #ICML #ICML2025