sam3-onnx: SAM3 Offline on Your GPU
Easily export and run SAM3 segmentation models with ONNX for cross-platform deployment.
Meta's SAM3 exported to ONNX, running on NVIDIA CUDA and Apple CoreML. Prompts can be text, example boxes, negative boxes or clicks. The author reports 0.37 s per text prompt on an M5 MacBook Pro and 0.08 s on an RTX 3080 Ti. Segmentation, local, no API.