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 2.42, ONNX Runtime 1.24.3	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:

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

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
StereoNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	196.394 ms	6 - 1359 MB	NPU
StereoNet	ONNX	float	Snapdragon® X2 Elite	180.442 ms	20 - 20 MB	NPU
StereoNet	ONNX	float	Snapdragon® X Elite	330.195 ms	19 - 19 MB	NPU
StereoNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	261.045 ms	6 - 1978 MB	NPU
StereoNet	ONNX	float	Qualcomm® QCS8550 (Proxy)	353.067 ms	0 - 24 MB	NPU
StereoNet	ONNX	float	Qualcomm® QCS9075	513.168 ms	3 - 6 MB	NPU
StereoNet	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	219.781 ms	3 - 1324 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	188.608 ms	3 - 3263 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® X2 Elite	194.982 ms	3 - 3 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® X Elite	362.226 ms	3 - 3 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	310.814 ms	3 - 4452 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	1293.97 ms	1 - 3260 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	433.689 ms	3 - 1112 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA8775P	461.879 ms	0 - 3260 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® QCS9075	510.602 ms	3 - 9 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA7255P	1293.97 ms	1 - 3260 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA8295P	515.878 ms	0 - 3366 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	237.626 ms	2 - 3245 MB	NPU
StereoNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	257.68 ms	72 - 3823 MB	NPU
StereoNet	TFLITE	float	Qualcomm® QCS9075	661.686 ms	72 - 202 MB	NPU
StereoNet	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	277.134 ms	73 - 3773 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.