We show that for human-object interaction detection a relatively simple factorized model with appearance and layout encodings constructed from pre-trained object detectors outperforms more ...sophisticated approaches. Our model includes factors for detection scores, human and object appearance, and coarse (box-pair configuration) and optionally fine-grained layout (human pose). We also develop training techniques that improve learning efficiency by: (1) eliminating a train-inference mismatch; (2) rejecting easy negatives during mini-batch training; and (3) using a ratio of negatives to positives that is two orders of magnitude larger than existing approaches. We conduct a thorough ablation study to understand the importance of different factors and training techniques using the challenging HICO-Det dataset.
The well-known correlation between structure and functionality has motivated generations of scientists to intensively investigate the structural behaviour of peptide and protein systems, e.g. their ...folding, their aggregation reactions or the process of molecular recognition. A variety of environmental effects on peptide structures occur among them the influence of solvent molecules or aggregation partners; a further decisive factor is the amino acid sequence. Thus a bottom-up approach comprises the investigation of isolated peptide systems, increasing in size, as well as a successive introduction of potential aggregation partners. For this purpose mass and isomer selective combined IR/UV investigations in a molecular expansion represent ideal experiments to analyse intrinsic structural properties of peptides and aggregates. Against this background the presented review article illustrates general aspects of peptide structure, spectroscopic methods and experimental set-ups in the first part. This overview is followed by a summary of the current results in this field of research including a more detailed discussion of our work but also selected findings of other groups.
We introduce SAIL-VOS (Semantic Amodal Instance Level Video Object Segmentation), a new dataset aiming to stimulate semantic amodal segmentation research. Humans can effortlessly recognize partially ...occluded objects and reliably estimate their spatial extent beyond the visible. However, few modern computer vision techniques are capable of reasoning about occluded parts of an object. This is partly due to the fact that very few image datasets and no video dataset exist which permit development of those methods. To address this issue, we present a synthetic dataset extracted from the photo-realistic game GTA-V. Each frame is accompanied with densely annotated, pixel-accurate visible and amodal segmentation masks with semantic labels. More than 1.8M objects are annotated resulting in 100 times more annotations than existing datasets. We demonstrate the challenges of the dataset by quantifying the performance of several baselines. Data and additional material is available at http://sailvos.web.illinois.edu.
Efficient Deep Learning for Stereo Matching Wenjie Luo; Schwing, Alexander G.; Urtasun, Raquel
2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR),
2016-June
Conference Proceeding
In the past year, convolutional neural networks have been shown to perform extremely well for stereo estimation. However, current architectures rely on siamese networks which exploit concatenation ...followed by further processing layers, requiring a minute of GPU computation per image pair. In contrast, in this paper we propose a matching network which is able to produce very accurate results in less than a second of GPU computation. Towards this goal, we exploit a product layer which simply computes the inner product between the two representations of a siamese architecture. We train our network by treating the problem as multi-class classification, where the classes are all possible disparities. This allows us to get calibrated scores, which result in much better matching performance when compared to existing approaches.
Changes in the amplitude and phasing of seasonal events (phenology) can affect the functioning of marine ecosystems. Phenology plays a particularly critical role in eastern boundary ecosystems, which ...are driven largely by the seasonal cycle of coastal upwelling. Here we develop and describe a set of indicators that quantify the timing, evolution, intensity, and duration of coastal upwelling in the California Current large marine ecosystem (CCLME). There is significant interannual variability in upwelling characteristics during 1967–2007, with extended periods of high (1970s, 1998–2004) and low (1980–1995) seasonally‐integrated upwelling and a trend towards a later and shorter upwelling season in the northern CCLME. El Niño years were characterized by delayed and weak upwelling in the central CCLME. Understanding the causes and ecosystem consequences of phenological changes in coastal upwelling is critical, as climate models project significant variability in the amplitude and phase of coastal upwelling under varying climate change scenarios.
In this paper we propose an approach to jointly estimate the layout of rooms as well as the clutter present in the scene using RGB-D data. Towards this goal, we propose an effective model that is ...able to exploit both depth and appearance features, which are complementary. Furthermore, our approach is efficient as we exploit the inherent decomposition of additive potentials. We demonstrate the effectiveness of our approach on the challenging NYU v2 dataset and show that employing depth reduces the layout error by 6% and the clutter estimation by 13%.
The goal of the Pacific Ocean Boundary Ecosystem and Climate Study (POBEX) was to diagnose the large-scale climate controls on regional transport dynamics and lower trophic marine ecosystem ...variability in Pacific Ocean boundary systems. An international team of collaborators shared observational and eddy-resolving modeling data sets collected in the Northeast Pacific, including the Gulf of Alaska (GOA) and the California Current System (CCS), the Humboldt or Peru-Chile Current System (PCCS), and the Kuroshio-Oyashio Extension (KOE) region. POBEX investigators found that a dominant fraction of decadal variability in basin- and regional-scale salinity, nutrients, chlorophyll, and zooplankton taxa is explained by a newly discovered pattern of ocean-climate variability dubbed the North Pacific Gyre Oscillation (NPGO) and the Pacific Decadal Oscillation (PDO). NPGO dynamics are driven by atmospheric variability in the North Pacific and capture the decadal expression of Central Pacific El Niños in the extratropics, much as the PDO captures the low-frequency expression of eastern Pacific El Niños. By combining hindcasts of eddy-resolving ocean models over the period 1950–2008 with model passive tracers and long-term observations (e.g., CalCOFI, Line-P, Newport Hydrographic Line, Odate Collection), POBEX showed that the PDO and the NPGO combine to control low-frequency upwelling and alongshore transport dynamics in the North Pacific sector, while the eastern Pacific El Niño dominates in the South Pacific. Although different climate modes have different regional expressions, changes in vertical transport (e.g., upwelling) were found to explain the dominant nutrient and phytoplankton variability in the CCS, GOA, and PCCS, while changes in alongshore transport forced much of the observed long-term change in zooplankton species composition in the KOE as well as in the northern and southern CCS. In contrast, cross-shelf transport dynamics were linked to mesoscale eddy activity, driven by regional-scale dynamics that are largely decoupled from variations associated with the large-scale climate modes. Preliminary findings suggest that mesoscale eddies play a key role in offshore transport of zooplankton and impact the life cycles of higher trophic levels (e.g., fish) in the CCS, PCCS, and GOA. Looking forward, POBEX results may guide the development of new modeling and observational strategies to establish mechanistic links among climate forcing, mesoscale circulation, and marine population dynamics.
Max-Sliced Wasserstein Distance and Its Use for GANs Deshpande, Ishan; Hu, Yuan-Ting; Sun, Ruoyu ...
2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),
2019-June
Conference Proceeding
Odprti dostop
Generative adversarial nets (GANs) and variational auto-encoders have significantly improved our distribution modeling capabilities, showing promise for dataset augmentation, image-to-image ...translation and feature learning. However, to model high-dimensional distributions, sequential training and stacked architectures are common, increasing the number of tunable hyper-parameters as well as the training time. Nonetheless, the sample complexity of the distance metrics remains one of the factors affecting GAN training. We first show that the recently proposed sliced Wasserstein distance has compelling sample complexity properties when compared to the Wasserstein distance. To further improve the sliced Wasserstein distance we then analyze its 'projection complexity' and develop the max-sliced Wasserstein distance which enjoys compelling sample complexity while reducing projection complexity, albeit necessitating a max estimation. We finally illustrate that the proposed distance trains GANs on high-dimensional images up to a resolution of 256×256 easily.
Newly developed high-speed, synchrotron-based X-ray computed microtomography enabled us to directly image pore-scale displacement events in porous rock in real time. Common approaches to modeling ...macroscopic fluid behavior are phenomenological, have many shortcomings, and lack consistent links to elementary porescale displacement processes, such as Haines jumps and snap-off. Unlike the common singular pore jump paradigm based on observations of restricted artificial capillaries, we found that Haines jumps typically cascade through 10-20 geometrically defined pores per event accounting for 64% of the energy dissipation. Real-time imaging provided a more detailed fundamental understanding of the elementary processes in porous media, such as hysteresis, snapoff, and ç on wetting phase entrapment, and it opens the way for a rigorous process for upscaling based on thermodynamic models.
In this paper we tackle the problem of instance-level segmentation and depth ordering from a single monocular image. Towards this goal, we take advantage of convolutional neural nets and train them ...to directly predict instance-level segmentations where the instance ID encodes the depth ordering within image patches. To provide a coherent single explanation of an image we develop a Markov random field which takes as input the predictions of convolutional neural nets applied at overlapping patches of different resolutions, as well as the output of a connected component algorithm. It aims to predict accurate instance-level segmentation and depth ordering. We demonstrate the effectiveness of our approach on the challenging KITTI benchmark and show good performance on both tasks.