Practical Stereo Matching via Cascaded Recurrent Network with Adaptive Correlation
Paper • 2203.11483 • Published
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® 8 Elite Gen 5 Mobile | 3976.252 ms | 53 - 72 MB | CPU |
| CREStereo | ONNX | float | Snapdragon® X2 Elite | 1931.597 ms | 181 - 181 MB | CPU |
| CREStereo | ONNX | float | Snapdragon® X Elite | 12549.903 ms | 182 - 182 MB | CPU |
| CREStereo | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 4570.802 ms | 36 - 59 MB | CPU |
| CREStereo | ONNX | float | Qualcomm® QCS8550 (Proxy) | 5518.426 ms | 38 - 41 MB | CPU |
| CREStereo | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 3873.392 ms | 20 - 36 MB | CPU |
| CREStereo | ONNX | float | Qualcomm® QCS9075 | 3845.649 ms | 71 - 73 MB | CPU |
| CREStereo | ONNX | float | Qualcomm® QCS8750 | 3873.392 ms | 20 - 36 MB | CPU |
| CREStereo | ONNX | float | Qualcomm® QCS7181 | 12549.903 ms | 182 - 182 MB | CPU |
| CREStereo | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 2842.539 ms | 153 - 231 MB | CPU |
| CREStereo | QNN_DLC | float | Snapdragon® X2 Elite | 2815.762 ms | 102 - 102 MB | CPU |
| CREStereo | QNN_DLC | float | Snapdragon® X Elite | 7447.066 ms | 105 - 105 MB | CPU |
| CREStereo | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 4543.495 ms | 140 - 221 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® QCS8275 | 11460.553 ms | 137 - 210 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 5906.09 ms | 142 - 174 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® SA8775P | 7133.704 ms | 137 - 210 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® SA8650P | 7133.704 ms | 137 - 210 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® SA8255P | 7133.704 ms | 137 - 210 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 7079.054 ms | 150 - 232 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® SA7255P | 11460.553 ms | 137 - 210 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® SA8295P | 5796.889 ms | 135 - 208 MB | CPU |
| CREStereo | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 3540.134 ms | 133 - 209 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® QCS9075 | 7388.887 ms | 306 - 1204 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® QCS8750 | 3540.134 ms | 133 - 209 MB | CPU |
| CREStereo | QNN_DLC | float | Qualcomm® QCS7181 | 7447.066 ms | 105 - 105 MB | CPU |
| CREStereo | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 4185.37 ms | 38 - 66 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® QCS8275 | 7101.187 ms | 41 - 65 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 5167.885 ms | 40 - 52 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® SA8775P | 5667.586 ms | 41 - 65 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® SA8650P | 5667.586 ms | 41 - 65 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® SA8255P | 5667.586 ms | 41 - 65 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 6975.385 ms | 41 - 71 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® SA7255P | 7101.187 ms | 41 - 65 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® SA8295P | 4419.139 ms | 40 - 64 MB | CPU |
| CREStereo | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 2931.315 ms | 37 - 64 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® QCS9075 | 5684.815 ms | 40 - 359 MB | CPU |
| CREStereo | TFLITE | float | Qualcomm® QCS8750 | 2931.315 ms | 37 - 64 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.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support