SAM3 ONNX models

SAM3 exported to ONNX by sam3-onnx. The models run with onnxruntime on CPU, NVIDIA CUDA and Apple CoreML.

File Size Input
sam3_image_encoder.onnx 1.9 GB image
sam3_language_encoder.onnx 1.6 GB text prompt
sam3_decoder.onnx 130 MB text and example boxes, including negative boxes
sam3_point_decoder.onnx 20 MB points

Run with osam

pip install 'osam[cpu]'  # macOS; use 'osam[gpu]' on NVIDIA
osam run sam3 --image cars.jpg --text car

Add --device coreml to run on the Apple silicon GPU.

Run with sam3-onnx

git clone https://github.com/wkentaro/sam3-onnx.git --recursive && cd sam3-onnx
uvx hf download wkentaro/sam3-onnx-models-v0.5.0 --local-dir models
uv run --no-dev infer_onnx.py --image images/bus.jpg --text-prompt person

License

The weights are under Meta's SAM License. The export code is MIT.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for wkentaro/sam3-onnx-models-v0.5.0

Base model

facebook/sam3
Quantized
(34)
this model