2018 MIT 6.S094 麻省理工深度学习和自动驾驶课程 中文
SegFuse: Dynamic Driving Scene Segmentation
DeepTesla - End-to-End Steering Model
相机 = 光测量装置(Camera = light-measuring device)
照明光源(Illumination source)(辐射(radiance)) -->
场景元素(Scene Element) --->
成像系统(Imaging System) --->
内部图像平面(Internal Image Plane) --->
输出(数字)图像(Output (digital) image)
图像 = 辐射能量测量(Image = radiant-energy measurement)
现代摄影流水线 Modern photography pipeline
场景辐射 ---> 相机前端(镜头过滤器 镜头Lens 快门Shutter 孔径) --->
相机内部(ccd响应response(RAW) CCD插值Demosaicing (原)) --->
相机后端处理(直方图均衡Hist equalization、空间扭曲Spatial warping)---> 输出
透过棱镜的白光 “White light” through a prism ------> 折射光(Refracted light)----> 光谱 Spectral
我们的眼睛有三个受体(锥细胞),它们对可见光作出反应并产生颜色感。
CSC320S: Introduction to Visual Computing 视觉计算导论
输入:17张raw图像,包括14张side images、2张top images、1张bottom image
输出:3D立体360度全景图像
深度摄影风格转换 Deep Photo Style Transfer
Self-augmented Convolutional Neural Networks
MeshStereo: A Global Stereo Model with Mesh Alignment Regularization for View Interpolation
1、讨论了立体视觉匹配(Stereo Matching)问题的定义及其与人眼感知深度的关系;
2、对Matching Cost Volume进行了可视化分析,以期望达到听者对其的直观且本质的理解;
3、讨论了立体视觉匹配问题中的四个经典方法(
Graph Cut,Adaptive Support Weight Aggregation,
Semi-Global Matching,
以及 PatchMatch Stereo);
4、讨论了MeshStereo的试图统一disparity求解以及网格生成两个步骤的motivation,
以及formulate这样一个unified model会遇到的困难;
5、讨论了MeshStereo引入splitting probability的解决方案及其优化过程。
Webinar最后展示了MeshStereo在深度估计以及新视角渲染两个任务中的结果。
Stereo Matching Using Tree Filtering non-local算法在双目立体匹配上的应用
Deep Learning Libraries 深度学习软件库
参考 人脸检测Face tracking and recognition database
视频分割 2010 ECCV Efficient Hierarchical Graph Based Video Segmentation
车辆检测 2002 ECCV Learning a sparse representation for object detection
参考 Caltech Pedestrian Detection Benchmark
Microsoft CoCo: Common Objects in Context
2012 ECCV Salient Objects Dataset (SOD)
2012 ECCV Neil D. B. Bruce Eye Tracking Data
2012 ECCV DOVES:A database of visual eye movements
2012 ECCV MSRA:Salient Object Database
2012 ECCV NUS: Predicting Saliency Beyond Pixels
2010 ECCV The DUT-OMRON Image Dataset
2010 ECCV An eye fixation database for saliency detection in images
UCF ChaoticInvariants datasetsActions
Hollywood Human Actions dataset data
Weizmann: Actionsas Space-Time Shapes
HMDB: A Large Video Database for Human Motion Recognition
MSR Action Recognition Datasets and Codes
Visual Event Recognition in Videos
2010 CVPR iCoseg: Interactive cosegmentation by touch
2010 CVPR Caltech-UCSD Birds 200
2009 ICCV An efficient algorithm for co-segmentation
2004 ECCV The Weizmann Horse Database
2013 BMVC Hierarchical Scene Annotation
2010 ECCV SuperParsing: Scalable Nonparametric Image Parsing with Superpixels
2009 CVPR Nonparametric Scene Parsing: Label Transfer via Dense Scene Alignment
2009 Scene Understanding Datasets
2008 IJCV The Daimler Urban Segmentation Dataset
2008 ECCV Motion-based Segmentation and Recognition Dataset)/CamVid/