YuNet face detection on the AMD NPU
An IRON export of https://github.com/opencv/opencv_zoo/tree/main/models/face_detection_yunet for AMD Ryzen AI NPUs: the compiled NPU kernels (.xclbin + instruction streams) and the packed weights an IRON Rust runtime replays.
The kernels are compiled for NPU2 (AIE2P: Strix Point, Strix Halo, Krackan) and will not load on NPU1 (Phoenix, Hawk Point).
Download
hf download brishen/iron-yunet-npu2 --local-dir yunet
# or, from an IRON checkout:
python scripts/hf_models.py download yunet --repo brishen/iron-yunet-npu2 --out yunet
Every convolution runs as an flm.GEMM on one hardware context (one xclbin, 36 instruction streams). manifest.txt and iron/applications/yunet/export_yunet.py describe the format; ref.* tensors hold one input and its reference outputs for a runtime's self-check.
Usage
See iron/applications/yunet/README.md in IRON for how the model runs on the NPU and how to point the runtime at this bundle.
Provenance
- Upstream model: https://github.com/opencv/opencv_zoo/tree/main/models/face_detection_yunet (license:
mit; its terms apply to these weights) - IRON commit:
44374ca - Uploaded: 2026-09-24
- Files: 40, 31.7 MB