Under submission
Troupe: Scaling Multi-Identity Generation to Crowds on a Frozen Few-Step Backbone
Duy A. Nguyen, Jitesh Chavan, Donghyun Lee, Mana Poustizadeh, Lucas Lorenzo Peña, Sam Huang, Liming Jiang, Khoa D. Doan, Jesse C. Boothe Rabek, Timo Mertens, Saurabh Shukla
Multi-identity personalization for 2 to 20 people from one reference photo each, using local identity conditioning on a frozen few-step backbone. Introduces the CrowdScale benchmark.
Under review
When Should Medical VLMs Answer? Reliability under Deployment Perturbations
Duy A. Nguyen, Tuan T. Nguyen, Anh-Tien Nguyen, Minh N. Do, Khoa D. Doan
How medical vision-language models behave under realistic, clinician-validated image corruptions, and when they should answer, ask for clarification, or abstain.
Conference · ACCV 2026 · Accepted
SURE: Scalable Uncertainty Estimation for Pretrained Multimodal Models with Missing Modalities
Duy A. Nguyen, Do Huu Quan, Khoa D. Doan, Minh N. Do
Workshop · ICML 2026, Efficient Multimodal Question Answering
Answer, Clarify, or Abstain: Fine-Grained Selective Prediction for Medical VLMs
Duy A. Nguyen, Tuan T. Nguyen, Minh N. Do, Khoa D. Doan