Newest papers

GazeLens for Efficient Egocentric Video Question Answering with Simple Gaze-Driven Adaptive Token Reduction

Trong-Thang Pham, Anh Nguyen, Hien Van Nguyen, Ngan Le · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28502 · cited by 0 works (OpenAlex) · open-access PDF

Indonesian

Kacamata cerdas dapat membantu pengguna memahami kegiatan dan lingkungan mereka, tetapi model bahasa penglihatan kecil harus mempertahankan bukti yang berguna dalam perhitungan terbatas. Ini sulit karena bingkai yang berlebihan mengkonsumsi anggaran visual, sementara kompresi ruang agresif dapat membuang detail yang relevan. Meskipun pandang mengidentifikasi fokus pemakai, pertanyaan mungkin memerlukan bukti dari adegan di sekitarnya. Untuk mengatasi keterbatasan ini, kami mengusulkan GazeLens, yang menggabungkan pemilihan bingkai dengan pengurangan token pandang dan pertanyaan. Pertama, transisi HOG dan ketajaman gambar memilih bingkai sebelum pengkodean visual. Untuk setiap bingkai yang dipilih, gabungan Gaussian kemudian membentuk ringkasan yang berfokus dan komplementer. Karena relevansi mereka tergantung pada pertanyaan, sebuah gerbang ringan menggabungkan mereka menjadi token primer. Bersama dengan skenario global,

Machine translation (Meta M2M100); the English original follows.

Abstract

Smart glasses can help users understand their activities and surroundings, but small vision-language models must preserve useful evidence within limited computation. This is difficult because redundant frames consume the visual budget, while aggressive spatial compression can discard relevant details. Although gaze identifies the wearer's focus, the question may require evidence from the surrounding scene. To address these constraints, we propose GazeLens, which combines frame selection with gaze- and question-guided token reduction. First, HOG transitions and image sharpness select frames before visual encoding. For each selected frame, Gaussian pooling then forms gaze-focused and complementary summaries. Because their relevance depends on the question, a lightweight gate blends them into a primary token. Together with a global scene token, this retains both focused and contextual summaries within two tokens per frame. On StreamGaze, GazeLens achieves higher macro and micro accuracy than the evaluated full-grid and token-reduction baselines, with lower input-forward FLOPs than the shared-backbone token-reduction baselines.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Security token, Pooling, Relevance (law), Artificial intelligence, Frame (networking)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange