Recent progress in computer vision has been driven by high-capacity models trained on large datasets. Unfortunately, creating large datasets with pixellevel labels has been extremely costly due to the amount of human effort required. In this paper, we present an approach to rapidly creating pixel-accurate semantic label maps for images extracted from modern computer games. Although the source code and the internal operation of commercial games are inaccessible, we show that associations between image patches can be reconstructed from the communication between the game and the graphics hardware. This enables rapid propagation of semantic labels within and across images synthesized by the game, with no access to the source code or the content. We validate the presented approach by producing dense pixel-level semantic annotations for 25 thousand images synthesized by a photorealistic open-world computer game. Experiments on semantic segmentation datasets show that using the acquired data to supplement real-world images significantly increases accuracy and that the acquired data enables reducing the amount of hand-labeled real-world data: models trained with game data and just 1 3 of the CamVid training set outperform models trained on the complete CamVid training set.

Figure 1: We provide evidence that state-of-the-art single-view 3D reconstruction methods (AtlasNet (light green, 0.38 IoU) [12], OGN (green, 0.46 IoU) [46], Matryoshka Networks (dark green, 0.47 IoU) [37]) do not actually perform reconstruction but image classification. We explicitly design pure recognition baselines (Clustering (light blue, 0.46 IoU) and Retrieval (dark blue, 0.57 IoU)) and show that they produce similar or better results both qualitatively and quantitatively. For reference, we show the ground truth (white) and a nearest neighbor from the training set (red, 0.76 IoU). The inset shows the input image. AbstractConvolutional networks for single-view object reconstruction have shown impressive performance and have become a popular subject of research. All existing techniques are united by the idea of having an encoder-decoder network that performs non-trivial reasoning about the 3D structure of the output space. In this work, we set up two alternative approaches that perform image classification and retrieval respectively. These simple baselines yield better results than state-of-the-art methods, both qualitatively and quantitatively. We show that encoder-decoder methods are statistically indistinguishable from these baselines, thus indicating that the current state of the art in single-view object reconstruction does not actually perform reconstruction but image classification. We identify aspects of popular experimental procedures that elicit this behavior and discuss ways to improve the current state of research.

Figure 1. Data for several tasks in our benchmark suite. Clockwise from top left: input video frame, semantic segmentation, semantic instance segmentation, 3D scene layout, visual odometry, optical flow. Each task is presented on a different image. AbstractWe present a benchmark suite for visual perception. The benchmark is based on more than 250K high-resolution video frames, all annotated with ground-truth data for both low-level and high-level vision tasks, including optical flow, semantic instance segmentation, object detection and tracking, object-level 3D scene layout, and visual odometry. Ground-truth data for all tasks is available for every frame. The data was collected while driving, riding, and walking a total of 184 kilometers in diverse ambient conditions in a realistic virtual world. To create the benchmark, we have developed a new approach to collecting ground-truth data from simulated worlds without access to their source code or content. We conduct statistical analyses that show that the composition of the scenes in the benchmark closely matches the composition of corresponding physical environments. The realism of the collected data is further validated via perceptual experiments. We analyze the performance of state-of-the-art methods for multiple tasks, providing reference baselines and highlighting challenges for future research. The supplementary video can be viewed at https://youtu.be/T9OybWv923Y

The diffuse very high−energy (VHE, > 100 GeV) γ-ray emission observed in the central 200 pc of the Milky Way by H.E.S.S. was found to follow the dense matter distribution in the Central Molecular Zone (CMZ) up to a longitudinal distance of about 130 pc to the Galactic Centre (GC), where the flux rapidly decreases. This was initially interpreted as the result of a burst−like injection of energetic particles 10 4 years ago, but a recent more sensitive H.E.S.S. analysis revealed that the cosmic−ray (CR) density profile drops with the distance to the centre, making data compatible with a steady cosmic PeVatron at the GC. In this paper, we extend this analysis to obtain for the first time a detailed characterisation of the correlation with matter and to search for additional features and individual γ-ray sources in the inner 200 pc. Taking advantage of 250 hours of H.E.S.S. data and improved analysis techniques we perform a detailed morphology study of the diffuse VHE emission observed from the GC ridge and reconstruct its total spectrum. To test the various contributions to the total γ-ray emission, we use an iterative 2D maximum likelihood approach that allows us to build a phenomenological model of the emission by summing a number of different spatial components. We show that the emission correlated with dense matter covers the full CMZ and that its flux is about half the total diffuse emission flux. We also detect some emission at higher latitude likely produced by hadronic collisions of CRs in less dense regions of the GC interstellar medium. We detect an additional emission component centred on the GC and extending over about 15 pc that is consistent with the existence of a strong CR density gradient and confirms the presence of a CR accelerator at the very centre of our Galaxy. We show that the spectrum of the full ridge diffuse emission is compatible with the one previously derived from the central regions, suggesting that a single population of particles fills the entire CMZ. Finally, we report the discovery of a VHE γ-ray source near the GC radio arc and argue that it is produced by the pulsar wind nebula candidate G0.13−0.11.

The IceCube neutrino observatory pursues a follow-up program selecting interesting neutrino events in real-time and issuing alerts for electromagnetic follow-up observations. In 2012 March, the most significant neutrino alert during the first three years of operation was issued by IceCube. In the follow-up observations performed by the Palomar Transient Factory (PTF), a Type IIn supernova (SN IIn) PTF12csy was found 0°. 2 away from the neutrino alert direction, with an error radius of 0°. 54. It has a redshift of z = 0.0684, corresponding to a luminosity distance of about 300 Mpc and the Pan-STARRS1 survey shows that its explosion time was at least 158 days (in host galaxy rest frame) before the neutrino alert, so that a causal connection is unlikely. The a posteriori significance of the chance detection of both the neutrinos and the SN at any epoch is 2.2σ within IceCubeʼs 2011/12 data acquisition season. Also, a complementary neutrino analysis reveals no long-term signal over the course of one year. Therefore, we consider the SN detection coincidental and the neutrinos uncorrelated to the SN. However, the SN is unusual and interesting by itself: it is luminous and energetic, bearing strong resemblance to the SN IIn