Arm/inception-v3-int8-mlas-onnx-raspberrypi5

Inception V3 optimized for Arm-based Edge Linux

Inception V3, a 1000-class ImageNet image classification model, optimized for Arm-based Edge Linux systems and delivered as an ONNX artifact running on ONNX Runtime.

Summary

This repository contains an Arm-optimized version of inception_v3 for image classification. The model is provided in ONNX running on the ONNX Runtime runtime, 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 ImageNet-1k and measured performance on a representative evaluation target.

The optimization is a static INT8 PTQ recipe: per-channel symmetric weights and per-tensor affine activations across the full network, with no layers excluded from quantization.

Key results

Area	Result
Model format	ONNX
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	113.11 ms p50 latency, 8.84 frames/sec
Accuracy result	77.45% Top-1, 93.52% Top-5 on ImageNet-1k
Size / memory result	23.71 MB, 3.83x smaller than the FP32 baseline

Original model

Field	Value
Original model	torchvision/inception_v3
Original source	torchvision/models/inception.py
Original developer	Google / PyTorch TorchVision
Original model card	torchvision.models.inception_v3
Original license	BSD-3-Clause

Model files

File	Description
inception_v3_raspberry_onnx_optimized.onnx	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 / accelerator	Cortex-A76 (4 cores @ 2.4 GHz), CPU execution
OS	Linux Raspberry Pi OS 64-bit, based on Debian 13 "Trixie"
Runtime	ONNX Runtime 1.28.0
Backend / delegate	MLAS
Batch size	1
Precision	INT8 static PTQ — per-channel symmetric weights, per-tensor affine activations
Runs	10 warmup + 100 measured

Performance results

Metric	Original / baseline	Arm-optimized	Improvement
p50 latency	406.76 ms	113.11 ms	3.60x faster
p90 latency	407.63 ms	113.30 ms	3.60x faster
p99 latency	415.85 ms	113.42 ms	3.67x faster
Throughput	2.46 frames/sec	8.84 frames/sec	3.59x
Model size	90.90 MB	23.71 MB	3.83x smaller
Peak memory	171.44 MB	123.94 MB	1.38x 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	ImageNet-1k
Split	val
Number of samples	50000
Metric(s)	Top-1, Top-5 accuracy
Evaluation runtime	ONNX Runtime

Accuracy results

Metric	Original / baseline	Arm-optimized	Change
Top-1 accuracy	77.51%	77.45%	-0.06 pp
Top-5 accuracy	93.57%	93.52%	-0.05 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 ONNX via the shared PT2E-to-ONNX pipeline
Quantization	Yes	INT8 static PTQ — per-channel symmetric weights, per-tensor affine activations across the full network, calibrated on 2000 randomly selected ImageNet-1k training samples
Runtime/backend selection	Yes	ONNX Runtime CPUExecutionProvider using MLAS kernels
Graph/runtime compatibility updates	Yes	Performed as part of the shared PT2E export pipeline, with ONNX-specific graph translation
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
Input shape	[1, 3, 299, 299]
Input type	float32
Input range	[0.0, 1.0]
Preprocessing	Resize the shorter edge to 342px, center-crop to 299x299, convert to a [0, 1] float32 tensor, then normalize with ImageNet mean/std

Expected output

Property	Value
Output shape	[1, 1000]
Output type	Raw class logits (unnormalized)
Postprocessing	Apply softmax over the 1000 logits and select the top-5 classes by probability

Intended use

This model is intended for developers evaluating image classification 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 ImageNet-1k (val, 50000 samples) 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

• Inference uses a fixed 299 x 299 input after resize and center crop, which differs from the 224 x 224 input used by most TorchVision classifiers.
• The auxiliary classification head (aux_logits) is a training-only branch, not part of standard inference, so it is disabled at export time and unavailable for inference.
• Model load time increased for the optimized model relative to FP32 (180.03 ms vs 102.32 ms), though steady-state latency is substantially faster.
• Sample input: sample_input.jpg is derived from Samoyed on Beach by Appleinfl, via Wikimedia Commons (public domain).

About this version

Original Model: torchvision/inception_v3 by Google / PyTorch TorchVision - 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 BSD-3-Clause.

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.