TanviKulkarni07/sam2-trt-engines

SAM 2.1 Hiera-Tiny ONNX Engines

This repository contains the optimized ONNX versions of the SAM 2.1 (Segment Anything Model 2) using the hiera_tiny backbone. These models have been exported with dynamic axes and device-agnostic tracing, making them ideal for high-performance inference in TensorRT or ONNX Runtime.

🚀 Model Components

To support the full video and image segmentation pipeline, the model is split into modular components:

File	Description
.onnx	Onnx models exported from Pytorch sam2 models with 'hiera_tiny' checkpoints.
.trt	TRT engines exported from Onnx models.

🔗 Video Segmentation

For real-time video segmentation using these TensorRT engines, visit my Github Repository.

Citations

If you use these weights, please cite the original SAM 2 research by Meta AI:

@inproceedings{ ravi2025sam, title={{SAM} 2: Segment Anything in Images and Videos}, author={Nikhila Ravi and Valentin Gabeur and Yuan-Ting Hu and Ronghang Hu and Chaitanya Ryali and Tengyu Ma and Haitham Khedr and Roman R{"a}dle and Chloe Rolland and Laura Gustafson and Eric Mintun and Junting Pan and Kalyan Vasudev Alwala and Nicolas Carion and Chao-Yuan Wu and Ross Girshick and Piotr Dollar and Christoph Feichtenhofer}, booktitle={The Thirteenth International Conference on Learning Representations}, year={2025}, url={https://openreview.net/forum?id=Ha6RTeWMd0} }