π¬ Holistic Video Scene Understanding
Understanding what is in the image or a video frame is crucial for many computer vision applications. Researchers from Google recently published a paper where they got state-of-the-art results in panoptic segmentation (see below for intuitive explanation). They achieved it by combining information about depth, existing and new objects from two consequent frames.
πSupplementary material:
β Read about the inverse projection problem here. βDo not scroll too much and try to guess what's in the image on the right π§
β To understand the difference between semantic, instance, and panoptic segmentation, see this image. If you would like to play with code and test panoptic segmentation yourself, visit the full article here.
π° Original article: Β Holistic Video Scene Understanding with ViP-DeepLab