Stereo Disparity Map Generation

Overview

State-of-the-art disparity map generation from 3D stereo images using PSMNet (Pyramid Stereo Matching Network). Completed as part of CMU’s Introduction to Deep Learning course.

Technical Details

• Model: PSMNet
• Framework: PyTorch
• Task: Stereo depth estimation
• Course: CMU Introduction to Deep Learning

Approach

Implemented and trained PSMNet to generate accurate disparity maps from stereo image pairs. The pyramid architecture enables multi-scale feature matching for robust depth estimation.

Results

Achieved state-of-the-art performance on benchmark stereo datasets through careful hyperparameter tuning and data augmentation strategies.