Computer Vision Architect
Autodesk Consumer Group
Adi Shavit holds an MSc. in Computer Science from The Hebrew University in Jerusalem, specializing in Computer Vision. His specialties include real-time image and video analysis and processing and the use and development of advanced machine learning algorithms for real world vision applications. Adi has worked extensively on porting and optimizing vision applications from embedded and mobile systems to multi-cluster cloud-based systems and holds several patents for technological innovations. Adi is a moderator and veteran contributor to the OpenCV open-source project. In Nov. 2012, Adi joined Autodesk with the rest of the CmyCasa team where he leads the Homestyler core algorithm development for cloud, web and mobile platforms.
Virtual Decluttering: Clean Your Room on Your Tablet!
Autodesk Homestyler is a free cross-platform mobile app that allows users to redecorate their home and other interior spaces.
Using a single photo of the space, advanced single image calibration and 3D reconstruction allows redecoration by inserting virtual furniture, re-painting and adding new flooring. Many user photos contain items like hanging pictures or old furniture that the user would like to remove or erase from the redecorated design. Removing these items exposes wall and flooring areas that were not visible in the original photo. These regions often contain complex and/or structured background textures such as wallpapers, rugs and tiles. Inpainting algorithms and content-aware-fill find such scenes difficult to process due to the detailed textures, significant perspective deformations and illumination differences in different parts of the image.
I will present Homestyler’s real-time interactive Conceal Tool which allows users to easily and seamlessly conceal image items while preserving perspective and illumination. The tool focuses on supporting the designer’s creative freedom while helping maintain a realistic and credible result. The algorithm takes into account Weber’s law of human perception to automatically determine the desired blending properties. The tool runs in realtime on most iOS and Android mobile phones and tablets.