11 Computer Vision News Qualitative comparisons to a traditional depth regression baseline for occlusion estimation. Jamie’s paper and associated YouTube video demonstrate that this method effectively handles moving objects despite being trained solely on static scenes, representing a significant achievement. However, the work relies on an indoor dataset for indoor 3D reconstruction and depth estimation, which is a potential limitation. A notable gap remains in the absence of an outdoor dataset from a phone capture point of view, featuring elements such as people, cars, and other objects in motion. If such a dataset were available, it would showcase the adaptability and efficacy of the method when confronted with moving objects, such as in scenarios where a character runs behind another moving person and occludes. Tackling this challenge would be a big step forward in the research. Virtual Occlusions Through Implicit Depth
RkJQdWJsaXNoZXIy NTc3NzU=