Pinned post
🚨 VideoAuto-R1 from Meta!🚨
🚀Excited to share our VideoAuto-R1: the first video understanding framework that only reasons when necessary. Our VideoAuto-R1 produces dramatically better results than video reasoning models such as Video-R1 by +10% on both video perception and reasoning benchmarks with significantly improved efficiency, reducing the average response length by∼3x. Our VideoAuto-R1 also outperforms Qwen3-VL-8B by >2.5% on video reasoning benchmarks. For those of you who want to know how we did it, I have 4 words for you: Thinking Once, Answering Twice. Check out the following for more details!
Paper: arxiv.org/abs/2601.05175
Project details and demo: ivul-kaust.github.io/projects/video…
with: @MingchenZhuge @zechunliu @garvinchen2 @liuzhuang1234 @Hu_Hsu @klightlm @ChongZhou7 @erniecyc @_sakshams_ @balakrishnan_vr @raghuraman @BernardSGhanem @vikasc @YoungXiong1 etc
🚀Excited to share our VideoAuto-R1: the first video understanding framework that only reasons when necessary. Our VideoAuto-R1 produces dramatically better results than video reasoning models such as Video-R1 by +10% on both video perception and reasoning benchmarks with significantly improved efficiency, reducing the average response length by∼3x. Our VideoAuto-R1 also outperforms Qwen3-VL-8B by >2.5% on video reasoning benchmarks. For those of you who want to know how we did it, I have 4 words for you: Thinking Once, Answering Twice. Check out the following for more details!
Paper: arxiv.org/abs/2601.05175
Project details and demo: ivul-kaust.github.io/projects/video…
with: @MingchenZhuge @zechunliu @garvinchen2 @liuzhuang1234 @Hu_Hsu @klightlm @ChongZhou7 @erniecyc @_sakshams_ @balakrishnan_vr @raghuraman @BernardSGhanem @vikasc @YoungXiong1 etc