ویراساینس.
سند علمی

PixelNet: Representation of the pixels, by the pixels, and for the pixels

Aayush Bansal, Xinlei Chen, Bryan C. Russell, Abhinav Gupta, Deva Ramanan

چکیده

We explore design principles for general pixel-level prediction problems, from low-level edge detection to midlevel surface normal estimation to high-level semantic segmentation. Convolutional predictors, such as the fullyconvolutional network (FCN), have achieved remarkable success by exploiting the spatial redundancy of neighboring pixels through convolutional processing. Though computationally efficient, we point out that such approaches are not statistically efficient during learning precisely because spatial redundancy limits the information learned from neighboring pixels. We demonstrate that stratified sampling of pixels allows one to (1) add diversity during batch updates, speeding up learning; (2) explore complex nonlinear predictors, improving accuracy; and (3) efficiently train state-of-the-art models tabula rasa (i.e., “from scratch”) for diverse pixel-labeling tasks. Our single architecture produces state-of-the-art results for semantic segmentation on PASCAL-Context dataset, surface normal estimation on NYUDv2 depth dataset, and edge detection on BSDS.

اسناد مشابه

  • PixelNet: Towards a General Pixel-level Architecture

    سند علمی۲۰۱۶CoRRAayush Bansal, Xinlei Chen, Bryan C. Russell, Abhinav Gupta

    We explore architectures for general pixel-level prediction problems, from low-level edge detection to mid-level surface normal estimation [4] to high-level semantic segmentation. Convolutional predictors, such as the fullyconvolutional network (FCN), have achieved remarkable suc

  • Pixel and Voxel Representations of Graphs

    سند علمی۲۰۱۵Graph DrawingMuhammad Jawaherul Alam, Thomas Bläsius, Ignaz Rutter, Torsten Ueckerdt

    We study contact representations for graphs, which we call pixel representations in 2D and voxel representations in 3D. Our representations are based on the unit square grid whose cells we call pixels in 2D and voxels in 3D. Two pixels are adjacent if they share an edge, two voxe

  • 1)local Feature Representations for Facial Expression Recognition Based on Differences of Gray Color Values of Neighboring Pixels

    سند علمی۲۰۱۳Mohammad Shahidul Islam

    Title of Dissertation Local Feature Representations for Facial Expression Recognition Based on Differences of Gray Color Values of Neighboring Pixels Author Mohammad Shahidul Islam Degree Doctor of Philosophy (Computer Science and Information Systems) Year 2013 A fruitful way for

  • Image Fusion for Spatial Enhancement of Hyperspectral Image via Pixel Group Based Non-Local Sparse Representation

    سند علمی۲۰۱۷Remote SensingJing Yang, Ying Li, Jonathan Cheung-Wai Chan, Qiang Shen

    Restricted by technical and budget constraints, hyperspectral images (HSIs) are usually obtained with low spatial resolution. In order to improve the spatial resolution of a given hyperspectral image, a new spatial and spectral image fusion approach via pixel group based non-loca

  • Pixel and semantic capabilities from an image-object based document representation

    سند علمی۲۰۰۷DRRMichael J. Gormish, Kathrin Berkner, Martin P. Boliek, Guotong Feng

    This paper reports on novel and traditional pixel and semantic operations using a recently standardized document representation called JPM. The JPM representation uses compressed pixel arrays for all visible elements on a page. Separate data containers called boxes provide the la

  • Beyond Pixels: Exploring New Representations and Applications for Motion Analysis

    سند علمی۲۰۰۹Edward H. Adelson

    The focus of motion analysis has been on estimating a flow vector for every pixel by matching intensities. In my thesis, I will explore motion representations beyond the pixel level and new applications to which these representations lead. I first focus on analyzing motion from v