转自: http://blog.csdn.net/scut1135/article/details/8012007 作者讲解视频: http://www.youtube.com/watch?v=_J_clwqQ4gI matlab代码实现: http://people.cs.uchicago.edu/~rbg/latent/ 开源C代码实现: https://github.com/liuliu/ccv ================================================================================== http://www.computervisiononline.com/software/discriminatively-trained-deformable-part-models Discriminatively Trained Deformable Part Models Over the past few years a complete learning-based system for detecting and localizing objects in images has been developed that represents objects using mixtures of deformable part models. These models are trained using a discriminative method that only requires bounding boxes for the objects in an image. The approach leads to efficient object detectors that achieve state of the art results on the PASCAL and INRIA person datasets. At a high level the system can be characterized by the combination of Strong low-level features based on histograms of oriented gradients (HOG) Efficient matching algorithms for deformable part-based models (pictorial structures) Discriminative learning with latent variables (latent SVM) This work was awarded the PASCAL VOC "Lifetime Achievement" Prize in 2010. Related publications: P. Felzenszwalb, D. McAllester, D. Ramaman , A Discriminatively Trained, Multiscale, Deformable Part Model IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2008 P. Felzenszwalb, R. Girshick, D. McAllester, D. Ramanan , Object Detection with Discriminatively Trained Part Based Models IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 32, No. 9, Sep. 2010 pdf P. Felzenszwalb, R. Girshick, D. McAllester, Cascade Object Detection with Deformable Part Models IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2010 pdf P. Felzenszwalb, D. McAllester, Object Detection Grammars University of Chicago, Computer Science TR-2010-02, February 2010 pdf R. Girshick, P. Felzenszwalb, D. McAllester, Object Detection with Grammar Models , Neural Information Processing Systems (NIPS), 2011 pdf R. Girshick, From Rigid Templates to Grammars: Object Detection with Structured Models , Ph.D. dissertation, The University of Chicago, Apr. 2012 pdf Software URL: http://people.cs.uchicago.edu/~rbg/latent/ ============================================================================= 改进版: http://agamenon.tsc.uah.es/Personales/rlopez/data/pose-estimation/ Deformable Part Models Revisited: A Performance Evaluation for ObjectCategory Pose Estimation Abstract Deformable PartModels (DPMs) as introduced by Felzenszwalb et al. have shown remarkably goodresults for category-level object detection. In this paper, we explore whetherthey are also well suited for the related problem of category-level object poseestimation. To this end, we extend the original DPM so as to improve itsaccuracy in object category pose estimation and design novel and more effectivelearning strategies. We benchmark the methods using various publicly availabledata sets. Provided that the training data is sufficiently balanced and clean,our method outperforms the state-of-the-art. Results Downloads Software Download a ready-to-use mDPM car pose estimator for 4, 8 a 16 views. Data Download the PASCAL VOC Augmented datasets (2006, 2007 and 2010) . Download the ICARO dataset . You may downloadthe paper too: DeformablePart Models Revisited: A Performance Evaluation for Object Category PoseEstimation , R. J. López-Sastre, T. Tuytelaars, S. Savarese. ICCV 2011 - 1stIEEE Workshop on Challenges and Opportunities in Robot Perception. Last update (11-Jan-12) with new results. ==================================================== 实验记录-deformable part models训练数据准备及中间结果记录 准备数据集: 1.生成新的imagelist 2.改变SelectImagesParsingXML()中if objectsize /imagesize = 0.9这一句的参数来选择bounding box比例不一样的图像,保存在imagelist中。 3.运行上述程序。 4.在VOCdevkit中新建文件夹,如VOC0002,并在其中添加5个文件夹,分别是: Annotations,ImageSets,JPEGImages,SegmentationClass,SegmentationObject。 5.修改SelectImagesToDirectory()函数中保存文件的位置,将上述imagelist中的图像及其标注复制到对应的文件夹中。有两个地方需要改动,如改成VOC0002。 还要改变对应的imagelistDIR = ;的imagelist的文件名字。这里表示imagelist中阈值设置为0.95时有201幅图像被选出。 7.准备trainval.txt,和train和val.txt。(注意应提前复制和准备好空文件,存放在experiment/VOCdevkit/VOC2002/ImageSets/Main文件夹下。 6.运行上述程序。程序的第一部分将image和annotation复制到文件夹中,第二部分生成对应的trainval.txt。 7.得到trainval.txt后,使用SplitTrainVal()将其随机分成两个集合,train和val。修改文件夹中的VOC0002即可。运行上述程序。 8.数据准备完毕,进入训练程序文件夹:~/experiment/mycode/voc-release4.01,比如说要运行pascal(‘person’,3);那么还需要改几个地方: 1.globals.m中的 VOCyear = '0002';需要修改 2.由于pascal_init.m中: tmp = pwd; cd(VOCdevkit); addpath( ); VOCinit; cd(tmp); 运行了VOCdevkit中VOCcode里的VOCinit,而其中包含了解析VOCopts的重要初始化内容,如果不改变会影响到后面的训练完全错误。 需要将VOCopts.dataset = 'VOC0001';中文件夹进行指定修改。 9.运行pascal(‘person’,3);注意应到mycode/voc-release4.01这个目录下。 运行1: 2. 3.阈值0.95,共201幅图像进行训练。
Centre for Research on the Epidemiology of Disasters http://www.emdat.be/ (国际灾害统计数据库) Since 1988 the WHO Collaborating Centre for Research on the Epidemiology of Disasters (CRED) has been maintaining an Emergency Events Database EM-DAT. EM-DAT was created with the initial support of the WHO and the Belgian Government. The main objective of the database is to serve the purposes of humanitarian action at national and international levels. It is an initiative aimed to rationalise decision making for disaster preparedness, as well as providing an objective base for vulnerability assessment and priority setting. EM-DAT contains essential core data on the occurrence and effects of over 18,000 mass disasters in the world from 1900 to present. The database is compiled from various sources, including UN agencies, non-governmental organisations, insurance companies, research institutes and press agencies. Top 10 most important Drought disasters for the period 1900 to 2011 sorted by numbers of total affected people at the country level: Country Date No Total Affected India, Drought May-1987 300,000,000 India, Drought Jul-2002 300,000,000 India, Drought 1972 200,000,000 India, Drought 1965 100,000,000 India, Drought Jun-1982 100,000,000 China P Rep, Drought Jan-1994 82,000,000 China P Rep, Drought Apr-2002 60,000,000 China P Rep, Drought Oct-2009 60,000,000 India, Drought Apr-2000 50,000,000 China P Rep, Drought Jun-1988 49,000,000 Top 10 most important Drought disasters for the period 1900 to 2011 sorted by numbers of killed at the country level: Country Date No Killed China P Rep, Drought 1928 3,000,000 Bangladesh, Drought 1943 1,900,000 India, Drought 1942 1,500,000 India, Drought 1965 1,500,000 India, Drought 1900 1,250,000 Soviet Union, Drought 1921 1,200,000 China P Rep, Drought 1920 500,000 Ethiopia, Drought May-1983 300,000 Sudan, Drought Apr-1983 150,000 Ethiopia, Drought Dec-1973 100,000