vietanhdev/segment-anything-2.1-onnx-models

Segment Anything 2.1 (SAM 2.1) — ONNX Models

ONNX exports of Meta's SAM 2.1 backbones — the maintenance release of SAM 2 — packaged for direct use with onnxruntime and AnyLabeling.

Why this repo exists

SAM 2.1 ships incremental quality improvements over SAM 2 with the same architecture and runtime cost. ONNX gives you a portable, dependency-light runtime that works in Python, C++, JavaScript, and most embedded targets. These exports are the ones AnyLabeling consumes for its smart-labeling features.

Variants

Each .zip bundles the encoder + decoder ONNX files for that backbone. The _20260221 suffix is the export date.

File	Backbone	Size
sam2.1_hiera_tiny_20260221.zip	Hiera-T	111 MB
sam2.1_hiera_small_20260221.zip	Hiera-S	136 MB
sam2.1_hiera_base_plus_20260221.zip	Hiera-B+	259 MB
sam2.1_hiera_large_20260221.zip	Hiera-L	768 MB

Quick start

pip install huggingface_hub onnxruntime

from huggingface_hub import hf_hub_download
import zipfile, onnxruntime as ort

zip_path = hf_hub_download(repo_id="vietanhdev/segment-anything-2.1-onnx-models",
                           filename="sam2.1_hiera_tiny_20260221.zip")
with zipfile.ZipFile(zip_path) as z:
    z.extractall("./sam21_tiny")

enc = ort.InferenceSession("./sam21_tiny/encoder.onnx", providers=["CPUExecutionProvider"])
print([(i.name, i.shape, i.type) for i in enc.get_inputs()])

For the full image → mask pipeline (encoder + decoder + prompt handling), see how AnyLabeling wires it: https://github.com/vietanhdev/anylabeling

Use with AnyLabeling

These models drop into AnyLabeling's auto-labeling backend without conversion. See the AnyLabeling docs for the model-config wiring.

Source weights

Original SAM 2.1 weights and license: https://github.com/facebookresearch/sam2

This repo redistributes the same weights in ONNX format. License unchanged from upstream (Apache 2.0).

Citation

@misc{nguyen2026sam21_onnx,
  author = {Nguyen, Viet-Anh and {Neural Research Lab}},
  title  = {SAM 2.1 ONNX Models},
  year   = {2026},
  url    = {https://huggingface.co/vietanhdev/segment-anything-2.1-onnx-models}
}

For the underlying model, cite Meta's SAM 2 paper (SAM 2.1 is a release of the same model family):

@article{ravi2024sam2,
  title   = {SAM 2: Segment Anything in Images and Videos},
  author  = {Ravi, Nikhila and others},
  journal = {arXiv:2408.00714},
  year    = {2024}
}

Acknowledgments

Thanks to Meta AI Research for the SAM 2.x line. This repo packages their work for edge inference.