Presenting our research on robust vision-language learning @ WACV2026

Two of our papers were accepted at the Winter Applications of Computer Vision (WACV) 2026 conference. The first one, “Multimodal Adversarial Defense for Vision-Language Models by Leveraging One-To-Many Relationships” (link), pioneers defense strategies against multimodal adversarial attacks on vision-language models by leveraging the one-to-many relationships between images and texts. The second one, “Training-free Conditional Image Embedding Framework Leveraging Large Vision Language Models” (link), proposes DIOR, a training-free approach that prompts a large vision-language model to produce conditional image embeddings focused on a given textual condition.