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 1.25.0	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® X2 Elite	206.031 ms	211 - 211 MB	NPU
StereoNet	ONNX	float	Snapdragon® X Elite	379.645 ms	148 - 148 MB	NPU
StereoNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	299.865 ms	6 - 4386 MB	NPU
StereoNet	ONNX	float	Qualcomm® QCS8550 (Proxy)	385.442 ms	0 - 49 MB	NPU
StereoNet	ONNX	float	Snapdragon® 8 Elite Mobile	250.085 ms	3 - 3239 MB	NPU
StereoNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	198.913 ms	3 - 3295 MB	NPU
StereoNet	ONNX	float	Qualcomm® QCS9075	530.939 ms	3 - 48 MB	NPU
StereoNet	ONNX	float	Qualcomm® QCS8750	250.085 ms	3 - 3239 MB	NPU
StereoNet	ONNX	float	Qualcomm® QCS7181	379.645 ms	148 - 148 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® X2 Elite	193.151 ms	3 - 3 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® X Elite	366.078 ms	3 - 3 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	285.954 ms	3 - 4452 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® QCS8275	1294.103 ms	3 - 3262 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	444.187 ms	3 - 7 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Elite Mobile	238.254 ms	0 - 3240 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA8295P	515.727 ms	0 - 3366 MB	NPU
StereoNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	186.378 ms	3 - 3303 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® QCS9075	511.428 ms	5 - 11 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® SA7255P	1294.103 ms	3 - 3262 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® QCS8750	238.254 ms	0 - 3240 MB	NPU
StereoNet	QNN_DLC	float	Qualcomm® QCS7181	366.078 ms	3 - 3 MB	NPU
StereoNet	TFLITE	float	Qualcomm® SA8775P	5701.173 ms	2 - 33 MB	CPU
StereoNet	TFLITE	float	Qualcomm® SA8650P	5701.173 ms	2 - 33 MB	CPU
StereoNet	TFLITE	float	Qualcomm® SA8255P	5701.173 ms	2 - 33 MB	CPU
StereoNet	TFLITE	float	Snapdragon® 8 Elite Mobile	280.229 ms	74 - 3775 MB	NPU
StereoNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	275.317 ms	73 - 3857 MB	NPU
StereoNet	TFLITE	float	Qualcomm® QCS9075	661.282 ms	72 - 202 MB	NPU
StereoNet	TFLITE	float	Qualcomm® QCS8750	280.229 ms	74 - 3775 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.