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	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 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:

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

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
CREStereo	ONNX	float	Snapdragon® X2 Elite	89.031 ms	2 - 2 MB	NPU
CREStereo	ONNX	float	Snapdragon® X Elite	202.859 ms	21 - 21 MB	NPU
CREStereo	ONNX	float	Snapdragon® 8 Gen 3 Mobile	144.267 ms	3 - 1334 MB	NPU
CREStereo	ONNX	float	Snapdragon® 8 Gen 1 Mobile	329.61 ms	4 - 1366 MB	NPU
CREStereo	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	270.871 ms	2 - 6 MB	NPU
CREStereo	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	200.621 ms	0 - 27 MB	NPU
CREStereo	ONNX	float	Qualcomm® QCS8450	329.61 ms	4 - 1366 MB	NPU
CREStereo	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	269.687 ms	2 - 5 MB	NPU
CREStereo	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	202.859 ms	21 - 21 MB	NPU
CREStereo	ONNX	float	Qualcomm® Dragonwing™ Q-8750	111.226 ms	1 - 1111 MB	NPU
CREStereo	ONNX	float	Snapdragon® 8 Elite Mobile	111.226 ms	1 - 1111 MB	NPU
CREStereo	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	79.192 ms	1 - 835 MB	NPU
CREStereo	QNN_DLC	float	Snapdragon® X2 Elite	88.044 ms	2 - 2 MB	NPU
CREStereo	QNN_DLC	float	Snapdragon® X Elite	203.342 ms	2 - 2 MB	NPU
CREStereo	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	144.861 ms	2 - 1239 MB	NPU
CREStereo	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	327.415 ms	2 - 1290 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	270.982 ms	2 - 7 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	199.79 ms	3 - 7 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® SA8775P	269.909 ms	3 - 1059 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® SA8650P	269.909 ms	3 - 1059 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® SA8255P	269.909 ms	3 - 1059 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® QCS8450	327.415 ms	2 - 1290 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	269.134 ms	4 - 8 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	203.342 ms	2 - 2 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	111.584 ms	2 - 1045 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® SA7255P	797.47 ms	3 - 1015 MB	NPU
CREStereo	QNN_DLC	float	Qualcomm® SA8295P	311.64 ms	3 - 1066 MB	NPU
CREStereo	QNN_DLC	float	Snapdragon® 8 Elite Mobile	111.584 ms	2 - 1045 MB	NPU
CREStereo	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	79.134 ms	1 - 748 MB	NPU
CREStereo	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	151.588 ms	0 - 1671 MB	NPU
CREStereo	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	380.241 ms	1 - 1641 MB	NPU
CREStereo	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	278.441 ms	0 - 44 MB	NPU
CREStereo	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	214.229 ms	1 - 8 MB	NPU
CREStereo	TFLITE	float	Qualcomm® SA8775P	282.886 ms	1 - 1410 MB	NPU
CREStereo	TFLITE	float	Qualcomm® SA8650P	282.886 ms	1 - 1410 MB	NPU
CREStereo	TFLITE	float	Qualcomm® SA8255P	282.886 ms	1 - 1410 MB	NPU
CREStereo	TFLITE	float	Qualcomm® QCS8450	380.241 ms	1 - 1641 MB	NPU
CREStereo	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	281.032 ms	0 - 43 MB	NPU
CREStereo	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	121.343 ms	1 - 1379 MB	NPU
CREStereo	TFLITE	float	Qualcomm® SA7255P	809.7 ms	1 - 1402 MB	NPU
CREStereo	TFLITE	float	Qualcomm® SA8295P	350.415 ms	1 - 1382 MB	NPU
CREStereo	TFLITE	float	Snapdragon® 8 Elite Mobile	121.343 ms	1 - 1379 MB	NPU
CREStereo	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	88.077 ms	0 - 1115 MB	NPU

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.