MemorySAM: Memorize Modalities and Semantics with Segment Anything Model 2 for Multi-modal Semantic Segmentation
Chenfei Liao, Xu Zheng, Yuanhuiyi Lyu, Haiwei Xue, Yihong Cao, Jiawen Wang, Kailun Yang, Xuming Hu
TL;DR
MemorySAM tackles multi-modal semantic segmentation by reinterpreting paired modalities as a sequence of frames and applying SAM2's memory mechanism to learn modality-agnostic features. It introduces a training-only Semantic Prototype Memory Module (SPMM) based on prototypical learning to inject semantic knowledge, supported by a LoRA-based fine-tuning strategy for the image encoder. The method achieves state-of-the-art results on DELIVER (65.38% mIoU) and MCubeS (52.88% mIoU), with substantial improvements over prior MMSS approaches while keeping parameter efficiency. These contributions demonstrate that combining modality-agnostic memory with semantic prototypes yields robust, scalable MMSS in both real-world and synthetic settings.
Abstract
Research has focused on Multi-Modal Semantic Segmentation (MMSS), where pixel-wise predictions are derived from multiple visual modalities captured by diverse sensors. Recently, the large vision model, Segment Anything Model 2 (SAM2), has shown strong zero-shot segmentation performance on both images and videos. When extending SAM2 to MMSS, two issues arise: 1. How can SAM2 be adapted to multi-modal data? 2. How can SAM2 better understand semantics? Inspired by cross-frame correlation in videos, we propose to treat multi-modal data as a sequence of frames representing the same scene. Our key idea is to ''memorize'' the modality-agnostic information and 'memorize' the semantics related to the targeted scene. To achieve this, we apply SAM2's memory mechanisms across multi-modal data to capture modality-agnostic features. Meanwhile, to memorize the semantic knowledge, we propose a training-only Semantic Prototype Memory Module (SPMM) to store category-level prototypes across training for facilitating SAM2's transition from instance to semantic segmentation. A prototypical adaptation loss is imposed between global and local prototypes iteratively to align and refine SAM2's semantic understanding. Extensive experimental results demonstrate that our proposed MemorySAM outperforms SoTA methods by large margins on both synthetic and real-world benchmarks (65.38% on DELIVER, 52.88% on MCubeS). Source code will be made publicly available.