EMIP introduces a two-stream architecture to address camouflaged segmentation and optical flow estimation simultaneously. The model incorporates a frozen pre-trained optical flow fundamental model and utilizes an interactive prompting scheme inspired by emerging visual prompt learning. By integrating segmentation-to-motion and motion-to-segmentation prompts, EMIP achieves state-of-the-art results on popular VCOD benchmarks.
На другой язык
из исходного контента
arxiv.org
Дополнительные вопросы