Look Beyond Saliency: Low-Attention Guided Dual Encoding for Video Semantic Search
For video retrieval in crowded environments, this method offers a simple, training-free enhancement to existing encoders.
The paper addresses the challenge of video semantic search in crowded scenes, where standard visual encoders overlook background context. The proposed Inverse Attention Embedding mechanism, combined with traditional embeddings, improves recall without extra training.
Video semantic search in densely crowded scenes remains a challenging task due to visual encoders tendency to prioritize salient foreground regions while neglecting contextually important, background areas. We propose an Inverse Attention Embedding mechanism that explicitly captures and highlights these overlooked regions. By combining inverse attention embeddings with traditional visual embeddings, our method significantly enhances semantic retrieval performance without additional training. Initial experiments and ablation studies demonstrate promising improvements over existing approaches in recall for video semantic search in crowded environments.