PriorTR (ECCV 2026): training-free, prior-corrected visual token reduction for accelerating multimodal LLMs — image & video.
A workshop for collections of multi-modal LLM examples, samples, reference architecture and demos on Amazon SageMaker.
A two-step framework for extracting textual answers from egocentric videos via NLQ, combining VSLNet for segment localization and Video-LLaVA for efficient answer generation.
(Accepted: NeurIPS 2025 Workshop Mexico City 7HVU) AdCare-VLM: Leveraging Large Vision Language Models (LVLMs) to Monitor Long-Term Medication Adherence and Care
Two-stage pipeline on the Ego4D NLQ benchmark: finding the moment in a first-person video that answers a question, then answering it — VSLNet/VSLBase ablation for temporal localisation, Video-LLaVA for answer generation, scored against hand-written ground truth.
To associate your repository with the video-llava topic, visit your repo's landing page and select "manage topics."