Abstract
The advent of large models, also known as foundation models, hassignificantly transformed the AI research landscape, with models like SegmentAnything (SAM) achieving notable success in diverse image segmentationscenarios. Despite its advancements, SAM encountered limitations in handlingsome complex low-level segmentation tasks like camouflaged object and medicalimaging. In response, in 2023, we introduced SAM-Adapter, which demonstratedimproved performance on these challenging tasks. Now, with the release ofSegment Anything 2 (SAM2), a successor with enhanced architecture and a largertraining corpus, we reassess these challenges. This paper introducesSAM2-Adapter, the first adapter designed to overcome the persistent limitationsobserved in SAM2 and achieve new state-of-the-art (SOTA) results in specificdownstream tasks including medical image segmentation, camouflaged (concealed)object detection, and shadow detection. SAM2-Adapter builds on theSAM-Adapter's strengths, offering enhanced generalizability and composabilityfor diverse applications. We present extensive experimental resultsdemonstrating SAM2-Adapter's effectiveness. We show the potential and encouragethe research community to leverage the SAM2 model with our SAM2-Adapter forachieving superior segmentation outcomes. Code, pre-trained models, and dataprocessing protocols are available athttp://tianrun-chen.github.io/SAM-Adaptor/