PaperarXiv (Cornell University) · 2023
Faster Segment Anything: Towards Lightweight SAM for Mobile Applications
1page built on it
0more pages cite it
2narratives
1registry paper cites it
In the Atlas
ModelPrimary source of
MobileSAM
Lightweight SAM family — replaces SAM's heavy ViT-H image encoder (632M params, ~452 ms on a single GPU) with a distilled TinyViT encoder (5.78M params, ~8 ms), keeping SAM's prompt encoder + mask decoder frozen and unchanged; MobileSAMv2 adds an object-aware prompt sampler (YOLOv8-style detector → bounding-box prompts) that replaces SAM's 32×32 grid-prompt + NMS pipeline for the Segment-Everything task, cutting end-to-end latency from ≈1616 ms to ≈97 ms (>16×) at equivalent mask quality.
Appears in 2 narratives
Lineage
Papers in the Atlas registry only · one dot per paperBuilt upon by · 1
Builds on · 1
2023Segment AnythingKirillov, Mintun et al.