A Unified Transformer Framework for Group-based Segmentation: Co-Segmentation, Co-Saliency Detection and Video Salient Object Detection

Humans tend to mine objects by learning from a group of images or several frames video since we live in dynamic world. In the computer vision area, many researchers focus on co-segmentation (CoS), co-saliency detection (CoSD) and salient object (VSOD) discover co-occurrent objects. However, previous approaches design different networks for these similar tasks separately, they are difficult apply each other. Besides, fail take full advantage cues among inter- intra-feature within images. this paper, introduce unified framework tackle issues view, term as UFGS (Unified Framework Group-based Segmentation). Specifically, first transformer block, which views image feature patch token then captures their long-range dependencies through self-attention mechanism. This can help network excavate patch-structured similarities relevant Furthermore, propose an intra-MLP module produce self-mask enhance avoid partial activation. Extensive experiments four CoS benchmarks (PASCAL, iCoseg, Internet MSRC), three CoSD (Cosal2015, CoSOD3k, CocA) five VSOD (DAVIS $_{16}$ , FBMS, ViSal, SegV2 DAVSOD) show that our method outperforms other state-of-the-arts both accuracy speed using same architecture, reach 140 FPS real-time. Code is available at https://github.com/suyukun666/UFO