ECCV 2020 Daily - Monday

The paper addresses the harder task of reconstructing scenes consisting of spatial arrangements of multiple objects. As well as predicting the shape of individual objects at their depicted pose, the model predicts the class each object belongs to e.g. chair, table, etc. It reconstructs the scene with all objects and the camera at their relative pose; detects and resolves occlusions, hallucinating their missing parts; and ensures each point in the output 3D space is occupied by at most one object. The team found their method improved over the state-of-the-art single-object 3D reconstruction methods on both ShapeNet and Pix3D datasets . Before we finish, we are keen to ask Stefan, what is it like to work with Vittorio Ferrari ? “He is very inspiring,” he enthuses. “He gets technical details very fast and has a way with logic. He is also very good at making strategic decisions, so he can prioritise what matters for your paper.” 3 CoReNet: Stefan Popov 11 DAILY M o n d a y To find out more about this work, visit Stefan’s oral session [#1084] and Q&A today (Monday) at 16:00 (UTC +1).