logo资料库

A Year in Computer Vision (2016 + 2017).pdf

发布时间：2022-06-28 发布人：admin 分类：说明书 资料大小：25.44M 资料格式：pdf 举报 版权申诉
第1页 / 共59页
第2页 / 共59页
第3页 / 共59页
第4页 / 共59页
第5页 / 共59页
第6页 / 共59页
第7页 / 共59页
第8页 / 共59页
资料共59页，剩余部分请下载后查看
    A YEAR IN COMPUTER VISION
    DEFINITION OF COMPUTER VISION The automatic extraction, analysis and understanding of useful information from a single image or a sequence of images. — BMVA (The British Machine Vision Association)
    CLASSIFICATION / LOCALISATION ¡ Classification: Assign a label to the whole image ¡ Localisation: Output a bounding box around the object in the image. ¡ Single object in the image.
    IMAGENET LARGE SCALE VISUAL RECOGNITION CHALLENGE (ILSVRC) • Classification Error and Localisation error are 0.023 and 0.062 in ILSVRC2017.
    INTERESTING TAKEAWAYS FROM ILSVRC 2016 • Scene Classification: Label an image with certain scene like “greenhouse”, “stadium”, etc. • Hikvision won scene classification with 9% top-5 error with an ensemble of deep Inception-style networks and not-so-deep residuals networks. • Trimps-Soushen use ensemble for classification, including Inception, Inception-Resnet, RestNet and Wide Residual Networks. For Localisation, Faster R-CNN is used. • ResNeXt extends the original ResNet architecture.
    INTERESTING TAKEAWAYS FROM ILSVRC 2017 • WMW: Squeeze-and-Excitation (SE) Building Block
    INTERESTING TAKEAWAYS FROM ILSVRC 2017 • NUS-Qihoo-DPNs: Dual Path Network, ResNet + DenseNet • The residual path implicitly reuses features, but it is not good at exploring new features. In contrast the densely connected network keeps exploring new features but suffers from higher redundancy.
    OBJECT DETECTION ¡ Object Detection: Outputting bounding boxes and labels for each individual objects in an image. ¡ One of the persistent issues in Object Detection to be the detection of small objects.
    分享到：
    赞 收藏

    相关推荐

    资料库

    音视频

    共收录313份资料，累计4个分类，关注成员有19位，主要包括：图像识别,直播技术,编解码,图像处理

    热门标签

    图像识别 直播技术 编解码 图像处理

    最新资料