Arm/yolov8s-int8-xnnpack-executorch

YOLOv8s optimized for Arm-based Edge Linux

This repository provides an Arm-optimized version of YOLOv8s for object detection, exported to ExecuTorch .pte format for efficient INT8 inference on Arm-based Edge Linux systems.

Summary

This repository contains an Arm-optimized version of YOLOv8s for object detection. 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 the COCO 2017 val2017 split (5000 samples) and measured performance on a representative evaluation target.

Key results

Area	Result
Model format	ExecuTorch .pte
Target device class	Edge Linux
Reference device	Raspberry Pi 5 (Cortex-A76, Linux Raspberry Pi OS 64-bit, based on Debian 13 "Trixie")
Primary performance result	p50 latency 457.03 ms, 2.19 frames per second (2.23x faster than the FP32 baseline); 95.43 percent of exported operators delegate to XNNPACK
Accuracy result	mAP@0.5:0.95 of 43.86 percent (-0.45 pp vs. the FP32 baseline)
Size / memory result	Model size 13.24 MB (3.23x smaller); peak memory 90.28 MB (1.56x less)

Original model

Field	Value
Original model	yolov8s
Original source	GitHub
Original developer	Ultralytics
Original model card	https://huggingface.co/Ultralytics/YOLOv8
Original license	AGPL-3.0

Model files

File	Description
yolov8s_raspberry_executorch_optimized.pte	Arm-optimized model for deployment
example.py	Minimal inference example
pyproject.toml	Pinned runtime dependencies and deployment metadata
uv.lock	Locked dependency versions, sources, and hashes for uv sync --frozen
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 (Cortex-A76, Linux Raspberry Pi OS 64-bit, based on Debian 13 "Trixie")
CPU / accelerator	Cortex-A76, arm64, 4 cores at 2.4 GHz
OS	Linux, Raspberry Pi OS 64-bit, based on Debian 13 "Trixie"
Runtime	ExecuTorch 1.1.0
Backend / delegate	XNNPACK, KleidiAI
Batch size	1
Precision	INT8 symmetric, per-channel weights
Runs	10 warmup runs, 100 measured runs

Performance results

Metric	Original / baseline	Arm-optimized	Improvement
p50 latency	1018.84 ms	457.03 ms	2.23x faster
p90 latency	1019.59 ms	457.52 ms	2.23x faster
p99 latency	1021.21 ms	460.91 ms	2.22x faster
Frames per second	0.98	2.19	2.23x
Model load time	73.40 ms	25.75 ms	2.85x faster
Time to first inference	1046.32 ms	471.89 ms	2.22x faster
Model size	42.78 MB	13.24 MB	3.23x smaller
Peak memory	141.06 MB	90.28 MB	1.56x less
Average memory	140.34 MB	89.34 MB	1.57x less

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
Split	val2017
Number of samples	5000
Metric(s)	mAP@0.5, mAP@0.75, mAP@0.5:0.95
Evaluation runtime	ExecuTorch (Arm-optimized); PyTorch eager (FP32 baseline)

Accuracy results

Metric	Original / baseline	Arm-optimized	Change
mAP@0.5	60.70	60.39	-0.31 pp
mAP@0.75	47.97	47.49	-0.48 pp
mAP@0.5:0.95	44.31	43.86	-0.45 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	Converted to ExecuTorch .pte via torch.export and PT2E static quantization
Quantization	Yes	INT8 static post-training quantization with symmetric, per-channel weights, calibrated on a stratified subset of 2,000 COCO 2017 images. Detection-head prediction convolutions (cv[2,3].*.2), DFL decode, sigmoid, output concatenation, and the first convolution remain in FP32; internal head Conv-BN-SiLU blocks, backbone, and neck are quantized
Runtime/backend selection	Yes	XNNPACK backend with KleidiAI kernels; 95.43 percent of exported operators delegate to XNNPACK
Graph/runtime compatibility updates	Yes	Internal C2f and SPPF concatenations are rewritten as equivalent split convolutions before quantization. After conversion, forced Q/DQ pairs are removed from remaining concatenations and the detection-head output is rewired to preserve FP32 box/class-score data flow
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

uv python install
uv sync --frozen

Run the example

uv run example.py

Expected input

Property	Value
Input shape	[1, 3, 640, 640]
Input type	float32
Input range	[0.0, 1.0]
Preprocessing	Letterbox resize to 640x640 with a pad value of 0.4471, converted to a CHW RGB tensor; no ImageNet-style normalization is applied

Expected output

Property	Value
Output shape	[1, 84, N] (or [1, N, 84])
Output type	N/A
Postprocessing	4 xywh center box coordinates plus 80 class scores, no objectness; sigmoid applied to class scores only if values fall outside [0, 1]; detection score is the max class score; filtered at a confidence threshold (0.001 for the reported mAP; the bundled example uses 0.25 for readable output); boxes converted to xyxy and filtered with per-class non-maximum suppression at an IoU threshold of 0.7; surviving boxes mapped back to original image coordinates via the inverse letterbox transform

Intended use

This model is intended for developers evaluating object-detection workloads on Arm-based platforms. It is suitable as a reference implementation for benchmarking, prototyping, and integration exploration.

Limitations

• Performance depends on the target device, runtime version, backend/delegate support, memory configuration, and system load.
• Accuracy was evaluated on the COCO 2017 val2017 split and may not generalize to all domains.
• This release preserves the original model's intended task and behavior, but users should validate it for their own application, data, and deployment environment.
• This repository is not a replacement for the original model documentation.

Additional notes

• Sample input: sample_input.jpg is derived from Living room (Unsplash) by Jarosław Ceborski, via Wikimedia Commons (CC0 1.0).

About this version

Original Model: yolov8s by Ultralytics - 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 AGPL-3.0.

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.

Use of Ultralytics models

Ultralytics provides Ultralytics YOLO software and models available through the Arm AI Portal and/or Arm's Hugging Face Organization under the GNU Affero General Public License v3.0 ("AGPL-3.0"), unless you have entered into a separate written license agreement with Ultralytics. Accessing, downloading, or retraining these materials through Arm AI Portal does not grant you an Ultralytics Enterprise License or any other proprietary Ultralytics license.

AGPL-3.0 requires you to release the complete source code of any application that uses Ultralytics YOLO, including applications made available over a network, under the same license. If you are embedding YOLO in a commercial product, internal tool, or production deployment and cannot open-source your code, you need an Ultralytics Enterprise License.

You are responsible for determining which applies to your use. Terms and enterprise licensing options are available at ultralytics.com.