Computer vision and image processing algorithms are computationally intensive. With CUDA acceleration, applications can achieve interactive video frame-rate performance. Here we outline some of the work in the area of imaging and vision and point to some resources for developers. 

Technical Reports on using CUDA for Imaging & Vision

• • Segmentation

    • "CUDA cuts: Fast graph cuts on the GPU"

  • Feature Processing

  • Fast Scale Invariant Feature Detection and Matching on Programmable Graphics Hardware

  • Fast Gain-Adaptive KLT Tracking on the GPU

  • Stereo Imaging

  • Stereo depth with a Unified Architecture GPU

  • Machine Learning & Data Processing

  • Large-scale Deep Unsupervised Learning using Graphics Processors

  • Hardware Efficient Belief Propagation

  • Fast k nearest neighbor search using GPU

• Biologically Inspired Computer Vision

Core Software Kernels for Imaging and Vision on CUDA GPUs

• Level-Set segmentation with CUDA

• Video segmentation with CUDA

• Multiclass SVM implementation in CUDA

• Pedestrian Detection

  • SIFT (Scale Invariant Feature Transform)

  • Marten Bjorkman’s Implementation

  • SiftGPU

  • Optical Flow

  • Flowlib: Dense Optical Flow

  • Baysian Optical Flow

  • Libraries and collections

  • GPU4Vision

  • OpenVIDIA: Popular computer vision algorithms on CUDA including

    • • Stereo Vision

      • Convolutions, Sobel, RMS, Histograms, Threshold, etc

  • MinGPU: A minimum GPU library for Computer Vision

• • • NVPP: NVIDIA Performance Primitives (Early access: Focuses on image and video processing)