Qualcomm® AI HubAI Hub

CREStereo

Cascaded recurrent stereo matching network for high‑accuracy disparity estimation.

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

Not supported

This model is currently not supported on any IoT chipset.

To see performance metrics for this model on other chipsets, click the button below.

View for other chipsets
Model RepositoryHugging FaceResearch Paper

Technical Details

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

Applicable Scenarios

  • Robotics
  • AR/VR
  • Autonomous Driving
  • 3D Reconstruction

License

Model:APACHE-2.0

Supported IoT Devices

  • Arduino VENTUNO Q
  • Dragonwing IQ-8275 EVK
  • Dragonwing IQ-9075 EVK
  • Dragonwing IQ-X5121
  • Dragonwing IQ-X7181
  • Dragonwing Q-8750
  • QCS8550 (Proxy)

Supported IoT Chipsets

  • Qualcomm® QCS5121
  • Qualcomm® Dragonwing™ IQ-X7181
  • Qualcomm® Dragonwing™ IQ-8275
  • Qualcomm® Dragonwing™ QCS8550 (Proxy)
  • Qualcomm® Dragonwing™ Q-8750
  • Qualcomm® Dragonwing™ IQ-9075

Related Models

See all models
View details for the StereoNet model.

StereoNet

Real-time stereo matching network for disparity estimation.

View details for the Midas-V2 model.

Midas-V2

Deep Convolutional Neural Network model for depth estimation.

View details for the Depth-Anything model.

Depth-Anything

Deep Convolutional Neural Network model for depth estimation.

View details for the Depth-Anything-V2 model.

Depth-Anything-V2

Deep Convolutional Neural Network model for depth estimation.

Looking for more? See models created by industry leaders.

Discover Model Makers