Quan Zhao | ViLab https://vilab.team/author/quan-zhao/ Quan Zhao Hugo Blox Builder (https://hugoblox.com)en-usWed, 08 Jan 2025 00:00:00 +0000 https://vilab.team/media/icon_hu2896232876136423579.png Quan Zhao https://vilab.team/author/quan-zhao/ Semantic-aware late-stage supervised contrastive learning for fine-grained action recognition https://vilab.team/publication/semantic-aware-late-stage-supervised-contrastive-learning-fo/ Wed, 08 Jan 2025 00:00:00 +0000 https://vilab.team/publication/semantic-aware-late-stage-supervised-contrastive-learning-fo/ <p>本文针对细粒度动作识别中类间差异小、类内差异大的挑战，提出了一种语义感知的后期监督对比学习方法。该方法通过后期监督对比学习策略，有效减少了对比学习所需的训练轮次，降低了计算成本；同时引入语义距离建模，在调整特征表示时显式考虑细粒度动作之间的语义关系，从而提升判别能力。实验表明该方法在多个细粒度动作识别基准上取得了优越性能。</p> Semantic-enhanced point-box joint prompting for video object segmentation https://vilab.team/publication/semantic-enhanced-point-box-joint-prompting-for-video-object/ Sun, 27 Oct 2024 00:00:00 +0000 https://vilab.team/publication/semantic-enhanced-point-box-joint-prompting-for-video-object/ <p>本文提出基于SAM的语义增强点框联合提示框架SAM-SPB，用于视频对象分割。该框架通过点跟踪分支维持对象局部结构信息，并利用语义感知的基于记忆的框跟踪分支跨帧传播对象语义一致性，从而结合局部与全局线索实现鲁棒分割。在主流VOS基准上取得了领先性能，验证了点框联合提示相比仅用点提示的优势。</p>