Lite-HRNet-18 optimized for Arm-based Edge Linux devices
An INT8-quantized version of lite_hrnet18 for keypoint detection, exported to ExecuTorch (.pte) and optimized for Arm-based Edge Linux devices such as the Raspberry Pi 5.
Summary
This repository contains an Arm-optimized version of lite_hrnet18 for keypoint detection, quantized to INT8 via static post-training quantization (per-channel symmetric weights, per-tensor affine activations). The model is provided in ExecuTorch (.pte) format, targeting Edge Linux systems.
This version is intended to demonstrate efficient inference on Arm-based platforms while preserving the original model's intended behavior. Arm has evaluated this model on COCO 2017 (keypoints) and measured performance on Raspberry Pi 5.
View the model listing on the Arm AI Portal for full data and code examples.
Key results
| Area | Result |
|---|---|
| Model format | ExecuTorch (.pte) |
| Target device class | Edge Linux |
| Reference device | Raspberry Pi 5 (Cortex-A76, Raspberry Pi OS 64-bit, based on Debian 13 "Trixie") |
| Primary performance result | 155.20 ms p50 latency (1.27x faster) |
| Accuracy result | 66.33% mAP@0.5:0.95 |
| Size / memory result | 2.82 MB, peak memory 50.25 MB |
Original model
| Field | Value |
|---|---|
| Original model | lite_hrnet18 |
| Original source | GitHub |
| Original developer | HRNet |
| Original model card | HRNet/Lite-HRNet |
| Original license | Apache-2.0 |
Model files
| File | Description |
|---|---|
lite_hrnet18_executorch_optimized.pte |
Arm-optimized INT8 model for deployment |
example.py |
Minimal inference example |
pyproject.toml |
Pinned runtime dependencies for example.py, resolved with uv |
uv.lock |
Locked dependency resolution for pyproject.toml |
config.yaml |
Model I/O contract used by the example |
benchmarks/ |
FP32 baseline and Arm-optimized benchmark records |
Performance
Performance was measured on the reference configuration below. Results are intended to make the optimization reproducible but do not guarantee identical performance on every Arm-based system.
Reference configuration
| Field | Value |
|---|---|
| Device / platform | Raspberry Pi 5 |
| CPU | Cortex-A76 (4 cores @ 2.4GHz) |
| OS | Raspberry Pi OS 64-bit, based on Debian 13 "Trixie" |
| Runtime | ExecuTorch 1.1.0 |
| Execution backend | CPU (XNNPACK, KleidiAI) |
| Precision | INT8 static PTQ — per-channel symmetric weights, per-tensor affine activations |
| Batch size | 1 |
| Input resolution | 288x384 |
| Runs / warmup | 100 / 10 |
Performance results
| Metric | Original / baseline | Arm-optimized | Improvement |
|---|---|---|---|
| End-to-end latency p50 (ms) | 197.17 | 155.20 | 1.27x faster |
| End-to-end latency p90 (ms) | 198.43 | 155.57 | 1.28x faster |
| Peak memory (MB) | 48.36 | 50.25 | 1.04x higher |
| Frames per second | 5.07 | 6.44 | 1.27x |
Accuracy
Accuracy was evaluated using the same preprocessing, input resolution, and evaluation protocol described below. Where possible, the optimized model is compared against the original model under the same evaluation conditions.
Evaluation setup
| Field | Value |
|---|---|
| Dataset | COCO 2017 (keypoints) |
| Split | val2017 |
| Sample count | 5000 |
| Metric(s) | mAP@0.5:0.95, mAP@0.5, mAP@0.75 |
| Runtime | ExecuTorch 1.1.0 (CPU) |
Accuracy results
| Metric | Original / baseline | Arm-optimized | Change |
|---|---|---|---|
| mAP@0.5:0.95 | 67.62% | 66.33% | -1.29 pp |
| mAP@0.5 | 87.78% | 87.01% | -0.78 pp |
| mAP@0.75 | 75.00% | 73.87% | -1.13 pp |
Accuracy was measured using the evaluation setup described above. Users should re-evaluate the model on their own data before production use.
Arm optimization approach
Arm optimized this model for efficient inference on Arm-based platforms using a hardware-aware conversion and validation flow.
For this release, Arm used:
| Optimization area | Applied? | Notes |
|---|---|---|
| Model conversion | Yes | Exported to ExecuTorch (.pte) via the shared PT2E capture/prepare/convert pipeline. |
| Quantization | Yes | INT8 static PTQ — per-channel symmetric weights, per-tensor affine activations; calibrated on 1,000 randomly sampled COCO 2017 (keypoints) training crops. |
| Runtime/backend selection | Yes | XNNPACK with KleidiAI kernels on ExecuTorch's CPU backend. |
| Graph/runtime compatibility updates | Yes | Performed as part of the ExecuTorch export pipeline. |
| Accuracy validation | Yes | Compared against the original model or published baseline |
| Performance validation | Yes | Measured on the reference Arm platform |
The goal of this process is to improve deployment characteristics such as latency, memory use, model size, and runtime compatibility while preserving the model's intended behavior. Detailed conversion scripts, calibration configuration, or backend-specific implementation details may be provided separately where appropriate.
Using this model
Install dependencies
Dependencies are declared in pyproject.toml, which ships with this repository. Resolve and install them into a local virtual environment with uv:
uv python install
uv sync --frozen
Run the example
uv run example.py
Expected input
| Property | Value |
|---|---|
| Shape | [1, 3, 384, 288] |
| Dtype | float32 |
| Range | [0.0, 1.0] before normalization |
| Preprocessing | Affine-warp a person bounding box (COCO x, y, w, h) onto a 288x384 canvas (box padded by a factor of 1.25, aspect ratio locked to 288/384 before padding), convert to tensor, normalize with mean [0.485, 0.456, 0.406] and standard deviation [0.229, 0.224, 0.225] |
Expected output
| Property | Value |
|---|---|
| Shape | [1, 17, 96, 72] |
| Format | Per-keypoint heatmaps for 17 COCO keypoints, stride-4 relative to the 288x384 crop (no output boxes or scores) |
| Postprocessing | Per-channel argmax, refined by a +/-0.25px quarter-pixel shift toward the second-highest neighbor, rescaled to the 288x384 crop, then reprojected to the original image by inverting the crop's affine transform |
Intended use
This model is intended for developers evaluating keypoint detection workloads on Arm-based platforms. It is suitable as a reference implementation for benchmarking, prototyping, and integration exploration.
Limitations
- This model has no person detector of its own — it requires an externally supplied person bounding box for every crop it processes.
- Requires a fixed 288x384 person crop; a different aspect ratio is padded rather than stretched.
- Accuracy was evaluated on COCO 2017 val2017, 5,000 images, cropped with an external person detector's boxes, and may not generalize to all domains.
- Performance measurements are from Raspberry Pi 5 hardware; results may differ on other Arm devices.
Additional notes
The network's deepest multi-branch fusion stage and its entire output stage (the iterative multi-resolution head and the final keypoint-decoding convolutions) are kept out of INT8 quantization to hold accuracy, since these regions carry a heavily skewed, outlier-dominated activation distribution that a per-tensor INT8 scale cannot represent. Calibration used 1,000 randomly sampled person crops from the COCO 2017 (keypoints) training split.
Sample input: sample_input.jpg is derived from Walking Rugby by Back ache, via Wikimedia Commons (CC0 1.0).
About this version
Original Model: lite_hrnet18 by HRNet - Repository
Optimization/conversion: Arm-Optimized version for execution on Arm-based platforms.
Converted/optimized by: Arm
License: The Original Model and the Optimized Model are subject to Apache-2.0.
This repository contains a converted or optimized version of the Original Model (the “Optimized Model”). The Original Model has been converted or optimized as described above for execution on Arm-based platforms.
No retraining or fine-tuning of the Original Model was performed as part of the conversion or optimization. The conversion or optimization was not intended to change the Original Model’s behavior or intended use.
Original Model and Documentation
For information about the Original Model, including its development, training data, intended uses, limitations and other relevant information, please refer to the Original Model repository. Information in that repository was provided by the original developer or other third parties and, unless expressly stated otherwise, has not been independently verified by Arm.
Licenses and Third-Party Terms
Use of the Original Model and the Optimized Model is subject to the applicable licenses, usage restrictions and other terms identified above and in the relevant repositories. Publication of the Optimized Model does not grant any rights beyond those provided under the applicable license terms.
You are responsible for reviewing those terms and ensuring that your use of the Original Model and the Optimized Model is permitted.
Purpose of this Release
The Optimized Model is provided as a reference implementation to demonstrate and evaluate execution and performance on Arm-based systems. It is not a production-ready or supported solution.
Arm’s publication of the Optimized Model does not constitute an endorsement or certification of the Original Model or a representation that the Optimized Model is suitable for production use or any particular purpose.
To the fullest extent permitted by applicable law (i) the Optimized Model is provided “as is.” Arm makes no representations or warranties that the Original Model, the Optimized Model or their outputs are accurate, safe, secure, non-infringing, legally compliant, suitable for production use or fit for any particular purpose; and (ii) Arm will not be liable for any loss or damage arising from or in connection with the Optimized Model, its use or its outputs.
You are responsible for independently evaluating the Optimized Model, its outputs and its suitability for your intended use, including compliance with applicable legal, regulatory, safety and security requirements.
Arm does not commit to provide ongoing support, maintenance or updates for the Optimized Model. Any use of or reliance on the Optimized Model or its outputs is at your own risk.
- Downloads last month
- 11