REU WEEK III Malcolm Collins-Sibley Mentor: Shervin Ardeshir
PROJECT Cross-View Image Registration and Semantic Segmentation The goal is to use information from map and satellite images, and project them on the screen which the user is observing, in a way that the user can see semantic segments overlaid on the scene.
PROJECT Output Mockup
COMPLETED WORK Readings: “Geometric Image Parsing in Man-Made Environments” Olga Barinova et al “Recovering Surface Layout from an Image” Derek Hoiem et al “Recovering Occlusion Boundaries from a Single Image” Derek Hoeim et al “Entropy Rate Superpixel Segmentation” MY Liu et al
COMPLETED WORK Geometric Image Parsing Code
COMPLETED WORK Geometric Image Parsing Code
COMPLETED WORK Super-pixel Segmentation With 8 super-pixels
COMPLETED WORK Super-pixel Segmentation With 20 super-pixels
COMPLETED WORK Building Projection
COMPLETED WORK Building Projection
CURRENT WORK Within the Building Projection code: Building occlusion and self-occlusion Works when a building occludes another, but not when a building is occluding itself
CURRENT WORK Occlusion Handling BeforeAfter
CURRENT WORK Occlusion Handling Before After
CURRENT WORK Occlusion Handling BeforeAfter
CURRENT WORK Occlusion Handling BeforeAfter
THE NEXT STEP Understanding the occlusion handling code Making sure it is handling self-occlusions accurately Understanding the format of the output data in the line segments/horizon code Running the line segmentation code for all of the images in our dataset and saving all of the output variables in a structure Extracting the super pixels from images in the dataset and saving it in a structure Computing their pairwise similarities of the super pixels in terms of color and texture