Arm/midas-v2-int8-xnnpack-executorch-graviton-g4

MiDaS V2 optimized for Arm-based Cloud CPU systems

An INT8-quantized version of MiDaS V2 for monocular depth estimation, exported to ExecuTorch (.pte) and optimized for Arm-based Cloud CPU systems.

Summary

This repository contains an Arm-optimized version of MiDaS V2 for monocular depth estimation, quantized to INT8 via static post-training quantization (PTQ) with per-channel symmetric weights and per-tensor affine activations. The model is provided in ExecuTorch (.pte) format, targeting Cloud CPU systems.

This version is intended to demonstrate efficient inference on Arm-based platforms while preserving the original model's intended behavior. Arm evaluated the model on NYU Depth V2 and measured the published performance and memory results on a representative evaluation target.

Key results

Area	Result
Model format	ExecuTorch .pte
Target device class	Cloud CPU
Reference device	AWS Graviton G4 (Neoverse-V2, 16 cores, Ubuntu 24.04.4 LTS)
Primary performance result	p50 latency 104.911 ms, 9.53 FPS (4.34x faster than baseline)
Accuracy result	RMSE 0.5732 m, AbsRel 0.1167, delta1 (< 1.25) 86.51%
Size / memory result	100.548 MB (3.95x smaller), peak memory 209.5 MB (3.78x less)

Original model

Field	Value
Original model	MiDaS V2
Original source	GitHub
Original developer	Intel ISL
Original model card	isl-org/MiDaS
Original license	MIT

Model files

File	Description
midas_v2_graviton_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	AWS Graviton G4
CPU / accelerator	Neoverse-V2, 16 cores / CPU
OS	Linux Ubuntu 24.04.4 LTS
Runtime	ExecuTorch 1.1.0
Backend / delegate	XNNPACK, KleidiAI
Batch size	1
Precision	INT8, static PTQ — per-channel symmetric weights, per-tensor affine activations
Input resolution	384x288
Runs	10 warmup + 50 measured

Performance results

Metric	Original / baseline	Arm-optimized	Improvement
Model size (MB)	397.153	100.548	3.95x smaller
End-to-end latency p50 (ms)	454.974	104.911	4.34x faster
End-to-end latency p90 (ms)	455.146	105.033	4.33x faster
End-to-end latency p99 (ms)	458.044	105.186	4.35x faster
Model load time (ms)	290.69	109.45	2.66x faster
Time to first inference (ms)	482.219	131.574	3.67x faster
Peak memory (MB)	792	209.5	3.78x less
Frames per second	2.2	9.53	4.33x

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	NYU Depth V2
Split	Validation
Number of samples	654
Metric(s)	RMSE (m), AbsRel, delta1 (< 1.25)
Evaluation runtime	ExecuTorch 1.1.0, CPU with XNNPACK and KleidiAI

Accuracy results

Metric	Original / baseline	Arm-optimized	Change
AbsRel (lower is better)	0.1145	0.1167	+0.0022
RMSE, m (lower is better)	0.5658	0.5732	+0.0074
delta1 (< 1.25), % (higher is better)	87.02	86.51	-0.51 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	Loaded from torch.hub as the MiDaS variant and exported to ExecuTorch .pte via the PT2E export pipeline
Quantization	Yes	INT8 static PTQ with XNNPACKQuantizer — per-channel symmetric weights, per-tensor affine activations, no layers kept in FP32; calibrated on 100 randomly selected NYU Depth V2 samples
Runtime/backend selection	Yes	XNNPACK + KleidiAI delegate on the ExecuTorch 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

The ordinary command prints a finite depth summary and does not overwrite the committed expected results. To write new output files, pass an explicit directory:

uv run example.py --output-dir outputs

The script reads sample_input.jpg, runs inference with midas_v2_graviton_executorch_optimized.pte, and optionally writes:

• sample_output.png — colourised depth map at the original image resolution (brighter is closer)
• depth.json — input/output metadata and raw/normalised depth statistics

Expected input

Property	Value
Input shape	[1, 3, 288, 384]
Input type	float32
Input range	[0.0, 1.0] before normalization
Preprocessing	Resize the RGB image to 288x384 (H x W) with bilinear interpolation and no aspect-ratio preservation; convert to a [0, 1] float32 tensor; normalize with ImageNet mean [0.485, 0.456, 0.406] and std [0.229, 0.224, 0.225]; add the batch dimension.

Expected output

Property	Value
Output shape	[1, 288, 384]
Output type	float32
Postprocessing	Squeeze the singleton dimensions; apply per-sample min-max normalization to [0, 1]; optionally resample to the original image resolution and apply a colour map for visualization.

Intended use

This model is intended for developers evaluating monocular depth estimation 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 654-image NYU Depth V2 validation split, which contains indoor scenes, and may not generalize to all domains.
• The input size is fixed at 288x384 and images are resized without preserving aspect ratio; inputs whose aspect ratio is far from 4:3 will be distorted.
• The output is relative inverse depth, not metric depth. Converting it to metric depth requires a scale-and-shift calibration outside the model.
• 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: MiDaS V2 by Intel ISL - 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 MIT.

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.