Human Pose Estimation 0001 — OpenVINO FP16
This repository distributes Intel's human-pose-estimation-0001 FP16 OpenVINO
IR, as used by Advantech Intel VisionAI pose estimation. It is a multi-person
2D pose network using an OpenPose-style approach and MobileNet v1 features,
with up to 18 body keypoints per person. Advantech did not train the original model.
Files and interface
Download human-pose-estimation-0001.xml and human-pose-estimation-0001.bin
into the same directory, retaining their names. This is IR v11. The original
XML records OpenVINO 2023.0.0/custom-HEAD conversion metadata. Artifacts are
unchanged from the existing VisionAI payload, without reconversion. FP16
describes compressed stored weights; input/output tensors are F32.
| Tensor | Type | Shape | Purpose |
|---|---|---|---|
Input data |
F32 | [1, 3, 256, 456] (NCHW) |
BGR image |
Output Mconv7_stage2_L1 |
F32 | [1, 38, 32, 57] |
Part affinity fields |
Output Mconv7_stage2_L2 |
F32 | [1, 19, 32, 57] |
Keypoint heatmaps |
Use the corresponding pose decoder to assemble keypoints into person skeletons.
The conversion includes BGR mean/scaling preprocessing; avoid duplicating it.
Result-layer names may include /sink_port_0; tensor names above describe the
model outputs. The demo may reshape the input to preserve the image aspect ratio.
Local use
Requires the VisionAI C++ pose demo, OpenVINO 2026.3 runtime and suitable driver. From the directory containing the downloaded files:
sha256sum -c SHA256SUMS
export LD_LIBRARY_PATH=/opt/Advantech/EdgeAI/System/Intel/SDK/runtime/openvino_2026_3/lib
/opt/Advantech/EdgeAI/System/Intel/VisionAI/app/bin/human_pose_estimation_demo \
-m ./human-pose-estimation-0001.xml -at openpose -d CPU -i /path/to/input.mp4 -loop
GPU/NPU compatibility and performance depend on runtime and drivers. No new accuracy evaluation or guaranteed video FPS is claimed. This repository provides model files for local execution, not a hosted pose-estimation service.
Sources and license
- Intel Open Model Zoo documentation
- Model manifest, download sources and license
- Open Model Zoo license
The model manifest references the Open Model Zoo Apache-2.0 license. Its text
is retained in LICENSE; conversion metadata remains in the XML. SHA256SUMS
covers both model files.