AI RESEARCH

SparseSAM: Structured Sparsification of Activations in Segment Anything Models

arXiv CS.CV

ArXi:2605.17633v1 Announce Type: new The Segment Anything Model (SAM) achieves strong open-vocabulary segmentation, but its ViT-based image encoders dominate inference latency and memory. Existing activation compression methods, such as token merging, reduce the token length to process, yet