SegmentAnything-TensorRT
lastmod 2026-11-10created 2024-06-262 edits
SegmentAnything-TensorRT is an implementation that runs Meta's image segmentation model Segment Anything (SAM) on NVIDIA TensorRT to speed up inference. It covers converting SAM's image encoder and mask decoder into TensorRT engines to generate masks faster than the original PyTorch version, and serves as a reference project for using SAM in near real time on machines with a GPU. It can be used as material for checking the conversion procedure and performance comparisons.
Welcome to the TensorRT implementation of the "Segment Anything" model!