Mosaix made the most sense on a mobile device because that's the platform on which a majority of our everyday photos are taken and stored. Our goal was to implement a reasonable approach to transforming photos into a feature vector, apply this transformation efficiently to the entire Photo Library, and then find the minimum difference between the vector of a subsection of our reference photo and the vector of some photo from the photo library so that we can replace it. At this point, the application reassembles the selected image (into the composite photo) of the other images in the users' libraries, presenting the user with the composite as a final result. Our application is a Swift mobile app with a fairly simple user interface the user selects the desired image (referred to as the reference photo) and adjusts the grid size and quality. We demonstrated the app live on an iPhone, by taking an on-the-spot picture and having the application generate a composite mosaic of that image on the spot with both a naive and parallel implementation.A video processing flow which, given a video, processes the video into frames and applies the photo mosaic algorithm to each frame so that it can be stitched back together into a transformed video.While there are a few apps already that perform similar functions to Mosaix, they consume 30-40s per image to finish processing (depending on grid size and quality). Performance that significantly improves upon existing solutions in the app store.Pre-processing of the device's Photo Library in the background that reduces complexity and time of the generation of each individual mosaic, and loading values previously computed from file to decrease the time required for steady-state photo mosaic production.Naive and basic parallel photo selection algorithms based around universal protocols, with an iterative approach to building a robust, performant data structure behind a parallelized solution.
PHOTO MOSAIC APPLICATION FULL
Full graphic user interface with automatic photo library access, an in-app camera and photo selector, and the ability to save mosaics.
Our iOS app currently exhibits about 20x speedup over the leading photo mosaic applications available in the App Store.Ī reference and composite comparison generated by our parallel algorithm. Using the photos already in the device's Photo Library, Mosaix reconstructs the reference photo by selecting photos which match in color, shape, and contrast and placing them in a grid pattern representing the original image. We created an iOS application which uses the Metal framework to generate photo mosaics in parallel, on-demand.
0 kommentar(er)
