cmsc733.github.io

Course materials and notes for University of Maryland's class CMSC733: Computer Vision

Project 1

• Panorama Stitching
• Supervised Deep Homography
• Unsupervised Deep Homography

Project 2

• Face Swap using Delaunay Triangulation and TPS (Facial Landmarks detected using dlib)
• Current ideas for Deep Learning
  • Implement Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network to regress 3D mesh of the face and use TPS to warp face and finally poisson blending to blend seamlessly
  • Implement Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network to regress 3D mesh of the face then implement Fast Face-swap Using Convolutional Neural Networks based blending using style transfer

References for Deep Learning

• On Face Segmentation, Face Swapping, and Face Perception
• Fast Face-swap Using Convolutional Neural Networks
• HeadOn: Real-time Reenactment of Human Portrait Videos
• Photorealistic Facial Texture Inference Using Deep Neural Networks
• Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network
• Deep Video Portraits
• Deep Fakes

Project 3

• Structure from Motion
  • Feature Extraction and Matching
  • Estimate F (with RANSAC)
  • Estimate E from F (with RANSAC)
  • Check Cheirality condition and Triangulation
  • Linear PnP
  • Non Linear PnP
  • Bundle Adjustment
  • Plot Camera Poses and 3D Structure
• SfM Learner
  • Depth Scale Alignment using predicted pose
  • Plot Camera Poses and 3D Structure

Project 4

• Reconstruct 3D Scene using Point to Plane ICP on RGBD data
• Implement SharpMask? to obtain segmentation in 2D
• Fuse outputs of SharkMask in 3D?