Arm/yolov8s-seg-int8-xnnpack-executorch

yolov8s-seg optimized for Arm-based Edge Linux systems

An INT8-quantized version of yolov8s-seg for instance segmentation, exported to ExecuTorch (.pte) and optimized for Arm-based Edge Linux systems.

Summary

This repository contains an Arm-optimized version of yolov8s-seg for instance segmentation, quantized to INT8 via static PTQ — 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 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 773.70 ms (1.80x faster than baseline), 1.29 fps
Accuracy result	Segm mAP@0.5:0.95 35.18%, Segm mAP@0.5 56.32%
Size / memory result	19.19 MB (2.36x smaller), peak memory 103.41 MB

Original model

Field	Value
Original model	yolov8s-seg
Original source	GitHub
Original developer	Ultralytics
Original model card	Ultralytics/YOLOv8
Original license	AGPL-3.0

Model files

File	Description
yolov8s-seg_raspberry_executorch_optimized.pte	Arm-optimized 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 @ 2.4 GHz
System memory	8 GB
OS	Linux, Raspberry Pi OS 64-bit, based on Debian 13 "Trixie"
Runtime	ExecuTorch 1.1.0
Backend / delegate	XNNPACK, KleidiAI (CPU, 1 thread)
Batch size	1
Input resolution	640x640
Precision	INT8 static PTQ — per-channel symmetric weights, per-tensor affine activations
Runs / warmup	100 / 10

Performance results

Metric	Original / baseline	Arm-optimized	Improvement
Model size (MB)	45.26	19.19	2.36x smaller
End-to-end latency p50 (ms)	1390.54	773.70	1.80x faster
End-to-end latency p90 (ms)	1391.94	774.23	1.80x faster
End-to-end latency p99 (ms)	1395.78	775.93	1.80x faster
Model load time (ms)	84.74	36.48	2.32x faster
Time to first inference (ms)	1415.03	797.45	1.77x faster
Peak memory (MB)	147.70	103.41	1.43x less
Average memory (MB)	146.97	102.48	1.43x less
Frames per second	0.72	1.29	1.79x

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
Sample count	5000
Metric(s)	Segm mAP@0.5:0.95, Segm mAP@0.5, Segm mAP@0.75 (mask-IoU instance-segmentation mAP)
Evaluation runtime	ExecuTorch 1.1.0 (CPU, XNNPACK, KleidiAI)

Accuracy results

Metric	Original / baseline	Arm-optimized	Change
Segm mAP@0.5:0.95 (%)	36.46	35.18	-1.28 pp
Segm mAP@0.5 (%)	57.61	56.32	-1.29 pp
Segm mAP@0.75 (%)	38.65	37.41	-1.24 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 PT2E capture/prepare/convert and XNNPACK lowering; predictions and proto-masks were wrapped into a single combined output tensor (ExecuTorch requires single-tensor output), followed by post-quantization graph surgery to strip residual Q/DQ pairs
Quantization	Yes	INT8 static PTQ via XNNPACKQuantizer with a Histogram observer — per-channel symmetric weights, per-tensor affine activations; calibrated on 2,000 COCO 2017 images (random selection); the segmentation head (except its self-contained proto.cv2 block) and the first convolution were skipped from quantization to keep accuracy within acceptable ranges
Runtime/backend selection	Yes	XNNPACK + KleidiAI delegate, single-threaded (1 thread)
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, 640, 640]
Dtype	float32
Range	[0.0, 1.0]
Preprocessing	aspect-preserving letterbox resize to fit 640x640 (no upscaling), padded with gray value 114/255 centered on the canvas; no normalization

Expected output

Property	Value
Shape	[1, total_size] (single combined tensor; ExecuTorch requires single-tensor output)
Format	Packs predictions [1, 116, N] (4 xywh pixel-unit box + 80 sigmoid-activated COCO class scores + 32 mask coefficients) followed by prototype masks [1, 32, 160, 160]; a trailing 4-value metadata tail is unreliable after INT8 quantization and must be ignored
Postprocessing	split deterministically into predictions and proto masks, compute per-anchor score as max(class scores), filter below the confidence threshold (0.25), convert boxes from xywh to xyxy, apply class-aware NMS (IoU threshold 0.7), reconstruct masks via sigmoid(mask coefficients @ proto masks) cropped to box and thresholded at 0.5, upsample 160x160 to 640x640, then invert the letterbox transform to map boxes and masks back to the original image

Intended use

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

Limitations

• Evaluated on 5,000 images from COCO 2017 val2017 (the standard validation split; not the full deployment distribution).
• Fixed input size of 640x640, letterboxed — no free aspect ratio; very small or extreme-aspect-ratio images are padded rather than distorted, but still lose resolution relative to native size.
• Performance depends on the target device, runtime version, backend/delegate support, memory configuration, and system load.
• This repository is not a replacement for the original model documentation.

Additional notes

Quantization used PT2E static PTQ via XNNPACKQuantizer with a Histogram observer. The segmentation head (except its self-contained proto.cv2 block) and the first convolution were kept in FP32 and skipped from quantization to keep accuracy within acceptable ranges. Calibration used 2,000 randomly selected images from COCO 2017.

• 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-seg 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.