The paper for ApolloScape says "This task is an extension of semantic object parsing by jointly considering detection and segmentation. Specifically, we select 39,212 training images and 1907 testing images...". Does this mean frames were randomly selected from the videos to be annotated with instance segmentations, or were all frames in a given video annotated with instance segmentations?
The paper for ApolloScape says "This task is an extension of semantic object parsing by jointly considering detection and segmentation. Specifically, we select 39,212 training images and 1907 testing images...". Does this mean frames were randomly selected from the videos to be annotated with instance segmentations, or were all frames in a given video annotated with instance segmentations?
https://arxiv.org/pdf/1803.06184.pdf