CREStereo: Optimized for Qualcomm Devices

CREStereo (Cascaded Recurrent Network with Adaptive Correlation) is a CVPR 2022 Oral paper that achieves state-of-the-art stereo matching accuracy.

This is based on the implementation of CREStereo 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	ONNX Runtime 1.25.0	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
TFLITE	float	Universal		Download

For more device-specific assets and performance metrics, visit CREStereo 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 CREStereo on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.depth_estimation

Model Stats:

• Model checkpoint: CREStereo ETH3D pretrained (crestereo_eth3d.pt)
• Input: Rectified stereo pair — left and right RGB images
• Input resolution: 240x320
• Output: Disparity map
• Number of parameters: 5.43M
• Model size (float): 20.7 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
CREStereo	ONNX	float	Snapdragon® X2 Elite	1705.182 ms	184 - 184 MB	CPU
CREStereo	ONNX	float	Snapdragon® X Elite	13237.424 ms	182 - 182 MB	CPU
CREStereo	ONNX	float	Snapdragon® 8 Gen 3 Mobile	4818.812 ms	35 - 55 MB	CPU
CREStereo	ONNX	float	Snapdragon® 8 Gen 1 Mobile	5936.471 ms	62 - 88 MB	CPU
CREStereo	ONNX	float	Qualcomm® QCS8550 (Proxy)	6053.112 ms	38 - 42 MB	CPU
CREStereo	ONNX	float	Qualcomm® QCS8450	5936.471 ms	62 - 88 MB	CPU
CREStereo	ONNX	float	Snapdragon® 8 Elite Mobile	3866.958 ms	9 - 25 MB	CPU
CREStereo	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	3689.841 ms	22 - 39 MB	CPU
CREStereo	ONNX	float	Qualcomm® QCS9075	3843.146 ms	71 - 73 MB	CPU
CREStereo	ONNX	float	Qualcomm® QCS8750	3866.958 ms	9 - 25 MB	CPU
CREStereo	ONNX	float	Qualcomm® QCS7181	13237.424 ms	182 - 182 MB	CPU
CREStereo	QNN_DLC	float	Snapdragon® X2 Elite	2763.679 ms	106 - 106 MB	CPU
CREStereo	QNN_DLC	float	Snapdragon® X Elite	8026.947 ms	106 - 106 MB	CPU
CREStereo	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	4840.298 ms	124 - 202 MB	CPU
CREStereo	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	7045.308 ms	150 - 231 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® QCS8275	11552.702 ms	136 - 208 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	5647.429 ms	138 - 174 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® SA8775P	7101.249 ms	136 - 209 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® SA8650P	7101.249 ms	136 - 209 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® SA8255P	7101.249 ms	136 - 209 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® QCS8450	7045.308 ms	150 - 231 MB	CPU
CREStereo	QNN_DLC	float	Snapdragon® 8 Elite Mobile	3499.965 ms	126 - 203 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® SA8295P	5651.013 ms	143 - 215 MB	CPU
CREStereo	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	2882.637 ms	109 - 187 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® SA7255P	11552.702 ms	136 - 208 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® QCS9075	7306.935 ms	308 - 1207 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® QCS8750	3499.965 ms	126 - 203 MB	CPU
CREStereo	QNN_DLC	float	Qualcomm® QCS7181	8026.947 ms	106 - 106 MB	CPU
CREStereo	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	4466.995 ms	39 - 65 MB	CPU
CREStereo	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	6864.416 ms	41 - 73 MB	CPU
CREStereo	TFLITE	float	Qualcomm® QCS8275	7076.53 ms	41 - 65 MB	CPU
CREStereo	TFLITE	float	Qualcomm® QCS8550 (Proxy)	5054.19 ms	37 - 49 MB	CPU
CREStereo	TFLITE	float	Qualcomm® SA8775P	5630.969 ms	41 - 65 MB	CPU
CREStereo	TFLITE	float	Qualcomm® SA8650P	5630.969 ms	41 - 65 MB	CPU
CREStereo	TFLITE	float	Qualcomm® SA8255P	5630.969 ms	41 - 65 MB	CPU
CREStereo	TFLITE	float	Qualcomm® QCS8450	6864.416 ms	41 - 73 MB	CPU
CREStereo	TFLITE	float	Snapdragon® 8 Elite Mobile	2875.056 ms	39 - 66 MB	CPU
CREStereo	TFLITE	float	Qualcomm® SA8295P	3718.184 ms	40 - 64 MB	CPU
CREStereo	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1097.651 ms	30 - 60 MB	CPU
CREStereo	TFLITE	float	Qualcomm® SA7255P	7076.53 ms	41 - 65 MB	CPU
CREStereo	TFLITE	float	Qualcomm® QCS9075	5679.714 ms	40 - 358 MB	CPU
CREStereo	TFLITE	float	Qualcomm® QCS8750	2875.056 ms	39 - 66 MB	CPU

License

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

References

• Practical Stereo Matching via Cascaded Recurrent Network with Adaptive Correlation
• 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.