Abstract: The visual enrichment of digital terrain models with plausible synthetic detail requires the segmentation of aerial images into a suitable collection of categories. In this paper we present a complete pipeline for segmenting high-resolution aerial images into a user-defined set of categories distinguishing e.g. terrain, sand, snow, water, and different types of vegetation. This segmentation-for-synthesis problem implies that per-pixel categories must be established according to the algorithms chosen for rendering the synthetic detail. This precludes the definition of a universal set of labels and hinders the construction of large training sets. Since artists might choose to add new categories on the fly, the whole pipeline must be robust against unbalanced datasets, and fast on both training and inference. Under these constraints, we analyze the contribution of common per-pixel descriptors, and compare the performance of state-of-the-art supervised learning algorithms. We report the findings of two user studies. The first one was conducted to analyze human accuracy when manually labeling aerial images. The second user study compares detailed terrains built using different segmentation strategies, including official land cover maps. These studies demonstrate that our approach can be used to turn digital elevation models into fully-featured, detailed terrains with minimal authoring efforts.

Abstract: Deep video action recognition models have been highly successful in recent years but require large quantities of manually-annotated data, which are expensive and laborious to obtain. In this work, we investigate the generation of synthetic training data for video action recognition, as synthetic data have been successfully used to supervise models for a variety of other computer vision tasks. We propose an interpretable parametric generative model of human action videos that relies on procedural generation, physics models and other components of modern game engines. With this model we generate a diverse, realistic, and physically plausible dataset of human action videos, called PHAV for “Procedural Human Action Videos”. PHAV contains a total of 39,982 videos, with more than 1000 examples for each of 35 action categories. Our video generation approach is not limited to existing motion capture sequences: 14 of these 35 categories are procedurally-defined synthetic actions. In addition, each video is represented with 6 different data modalities, including RGB, optical flow and pixel-level semantic labels. These modalities are generated almost simultaneously using the Multiple Render Targets feature of modern GPUs. In order to leverage PHAV, we introduce a deep multi-task (i.e. that considers action classes from multiple datasets) representation learning architecture that is able to simultaneously learn from synthetic and real video datasets, even when their action categories differ. Our experiments on the UCF-101 and HMDB-51 benchmarks suggest that combining our large set of synthetic videos with small real-world datasets can boost recognition performance. Our approach also significantly outperforms video representations produced by fine-tuning state-of-the-art unsupervised generative models of videos.