Predicting Deeper into the Future of Semantic Segmentation
This addresses the need for real-time anticipation in robotics and autonomous driving by improving future scene understanding, though it is incremental as it builds on existing prediction methods by focusing on segmentations.
The paper tackles the problem of predicting future semantic segmentation maps from video sequences, introducing a novel task that directly forecasts segmentations up to a second ahead. Results on Cityscapes show this approach is substantially better than predicting RGB frames first and then segmenting, with visually convincing predictions up to half a second and much higher accuracy than an optical flow baseline.
The ability to predict and therefore to anticipate the future is an important attribute of intelligence. It is also of utmost importance in real-time systems, e.g. in robotics or autonomous driving, which depend on visual scene understanding for decision making. While prediction of the raw RGB pixel values in future video frames has been studied in previous work, here we introduce the novel task of predicting semantic segmentations of future frames. Given a sequence of video frames, our goal is to predict segmentation maps of not yet observed video frames that lie up to a second or further in the future. We develop an autoregressive convolutional neural network that learns to iteratively generate multiple frames. Our results on the Cityscapes dataset show that directly predicting future segmentations is substantially better than predicting and then segmenting future RGB frames. Prediction results up to half a second in the future are visually convincing and are much more accurate than those of a baseline based on warping semantic segmentations using optical flow.