StereoNet: Optimized for Qualcomm Devices

StereoNet is an end-to-end deep architecture for real-time stereo matching that produces high-quality, edge-preserved disparity maps from a rectified stereo image pair.

This is based on the implementation of StereoNet found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime	Precision	Chipset	SDK Versions	Download
ONNX	float	Universal	QAIRT 2.45, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download

For more device-specific assets and performance metrics, visit StereoNet on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

• Custom weights (e.g., fine-tuned checkpoints)
• Custom input shapes
• Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for StereoNet on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.depth_estimation

Model Stats:

• Input resolution: 786x490
• Model checkpoint: KeystoneDepth (epoch=21-step=696366.ckpt)
• Model size (float): 7.41 MB
• Number of parameters: 1.94M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
StereoNet	ONNX	float	Snapdragon® X2 Elite	207.049 ms	5 - 5 MB	NPU
StereoNet	ONNX	float	Snapdragon® X Elite	378.219 ms	44 - 44 MB	NPU
StereoNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	296.95 ms	6 - 4395 MB	NPU
StereoNet	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	495.508 ms	3 - 9 MB	NPU
StereoNet	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	429.401 ms	3 - 52 MB	NPU
StereoNet	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	464.01 ms	3 - 9 MB	NPU
StereoNet	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	378.219 ms	44 - 44 MB	NPU
StereoNet	ONNX	float	Qualcomm® Dragonwing™ Q-8750	250.438 ms	3 - 3236 MB	NPU
StereoNet	ONNX	float	Snapdragon® 8 Elite Mobile	250.438 ms	3 - 3236 MB	NPU
StereoNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	199.162 ms	3 - 3303 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® X2 Elite	193.328 ms	3 - 3 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® X Elite	365.661 ms	3 - 3 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	285.047 ms	3 - 4451 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	482.907 ms	3 - 10 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	473.14 ms	3 - 8 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA8775P	461.966 ms	1 - 3260 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA8650P	461.966 ms	1 - 3260 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA8255P	461.966 ms	1 - 3260 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	450.574 ms	3 - 9 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	365.661 ms	3 - 3 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	237.798 ms	0 - 3238 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA7255P	1293.634 ms	1 - 3259 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA8295P	516.242 ms	0 - 3366 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Elite Mobile	237.798 ms	0 - 3238 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	187.254 ms	3 - 3301 MB	NPU
StereoNet	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	551.746 ms	74 - 78 MB	NPU
StereoNet	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	276.155 ms	73 - 3773 MB	NPU
StereoNet	TFLITE	float	Snapdragon® 8 Elite Mobile	276.155 ms	73 - 3773 MB	NPU
StereoNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	288.841 ms	74 - 3864 MB	NPU

License

• The license for the original implementation of StereoNet can be found here.

References

• StereoNet: Guided Hierarchical Refinement for Real-Time Edge-Aware Depth Prediction
• Source Model Implementation

Community

• Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
• For questions or feedback please reach out to us.