Official Implementation for "SAVI: slope-aware sparse-view sinogram interpolation"

This project contains CT image reconstruction code utilizing Soft-DTW (Dynamic Time Warping) implemented using OpenAI Triton.

Key Features

• Triton-based Acceleration: Efficiently performs distance calculations and Soft-DTW operations on the GPU via euclidean_dist_triton.py and soft_dtw_triton.py.
• Sinogram Alignment & Reconstruction: Uses Soft-DTW to align sinograms, leading to improved reconstruction results.

Requirements

• PyTorch (we conducted all experiments using version 2.7.0)
• Triton (automatically installed when you install pytorch)
• LEAP (if using Projector)

Usage

Just run the DTW_demo_parallel.ipynb notebook!!

Our approach does not require any training steps! (plug-and-playable)

Acknowledgements

Our code uses LivermorE AI Projector for Computed Tomography (LEAP) framework to perform filtered back-projection (FBP) for reconstruction. We appreciate for their contributions to the CT imaging community!