Arm/mobilegaze-mobilenet-v2-int8-xnnpack-executorch

MobileGaze MobileNetV2 INT8 — Gaze Estimation

This model estimates face-level yaw and pitch from an RGB face crop. It is the MobileGaze MobileNetV2 checkpoint converted to an A8W8, statically quantized ExecuTorch program using the XNNPACK backend.

Artifact

Item	Value
File	l2cs_mobilegaze_mobilenetv2-a8w8-2-xnnpack.pte
SHA-256	eec432891b173384c8122988a0bebec004419fa9374db15d936844845ea2dd42
Size	2,714,856 bytes
Runtime	ExecuTorch 1.1.0, XNNPACK backend
Input	FLOAT [1, 3, 448, 448], RGB NCHW
Outputs	yaw logits [1, 90]; pitch logits [1, 90]

See config.yaml for the inference recipe.

Example

By default, the example runs all four face crops under samples/ and writes their predicted gaze-direction overlays under assets/.

Input	Inferred visualization

python -m pip install --extra-index-url https://download.pytorch.org/whl/cpu \
  executorch==1.1.0 torch==2.10.0 numpy pillow pyyaml
python example.py
python example.py samples/face-crop-1.png assets/custom-gaze.png

The script parses config.yaml, making its shapes, resize, normalization, output semantics, and angle-bin postprocessing an executable inference recipe. Image loading, preprocessing, ExecuTorch inference, output postprocessing, visualization, and saving remain separate steps for reuse.

With ExecuTorch 1.1.0 on x86-64 CPU, the bundled samples produced yaw/pitch angles of -35.02°/-51.10°, 13.56°/-25.04°, -53.37°/-10.85°, and 38.23°/3.43°, respectively. These are interface examples, not accuracy or performance measurements.

MobileGaze expects an already detected face crop. It does not perform face detection.

Download

hf download Arm/mobilegaze-mobilenet-v2-int8-xnnpack-executorch \
  l2cs_mobilegaze_mobilenetv2-a8w8-2-xnnpack.pte \
  --local-dir .

Intended use and limitations

• Intended for yaw/pitch estimation on already detected, tightly cropped faces.
• It is not a face detector and does not identify what object a person is viewing.
• Performance may vary with pose, occlusion, lighting, crop quality, and camera placement.

References

• MobileGaze source repository
• L2CS-Net source repository
• MobileGaze publication record

About this version

Original Model: MobileGaze MobileNetV2 by Yakhyokhuja Valikhujaev - Repository

Optimization/conversion: Arm-Optimized version for execution on Arm-based platforms.

Converted/optimized by: Arm

License: The published artifacts in this repository are distributed under 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.