机器视觉开源代码集合
一、特征提取FeatureExtraction：
§ SIFT [1] [][] []
§ PCA-SIFT[2] []
§ Affine-SIFT[3] []
§ SURF [4] [] []
§ AffineCovariant Features [5] []
§ MSER [6] [] []
§ GeometricBlur [7] []
§ LocalSelf-Similarity Descriptor [8] []
§ Global andEfficient Self-Similarity [9] []
§ Histogramof Oriented Graidents [10] [] []
§ GIST [11][]
§ ShapeContext [12] []
§ ColorDescriptor [13] []
§ Pyramidsof Histograms of Oriented Gradients []
§ Space-TimeInterest Points (STIP) [14][] []
§ BoundaryPreserving Dense Local Regions [15][]
§ WeightedHistogram[]
§ Histogram-basedInterest Points Detectors[][]
§ An OpenCV- C++ implementation of Local Self Similarity Descriptors []
§ FastSparse Representation with Prototypes[]
§ CornerDetection []
§ AGASTCorner Detector: faster than FAST and even FAST-ER[]
§ Real-timeFacial Feature Detection using Conditional Regression Forests[]
§ Global andEfficient Self-Similarity for Object Classification and Detection[]
§ WαSH:Weighted α-Shapes for Local Feature Detection[]
§ HOG[]
§ OnlineSelection of Discriminative Tracking Features[]
二、图像分割ImageSegmentation：
§ NormalizedCut [1] []
§ Gerg Mori’Superpixel code [2] []
§ EfficientGraph-based Image Segmentation [3] [] []
§ Mean-ShiftImage Segmentation [4] [] []
§ OWT-UCMHierarchical Segmentation [5] []
§ Turbepixels[6] [] [] []
§ Quick-Shift[7] []
§ SLICSuperpixels [8] []
§ Segmentationby Minimum Code Length [9] []
§ BiasedNormalized Cut [10] []
§ SegmentationTree [11-12] []
§ EntropyRate Superpixel Segmentation [13] []
§ FastApproximate Energy Minimization via Graph Cuts[][]
§ EfficientPlanar Graph Cuts with Applications in Computer Vision[][]
§ IsoperimetricGraph Partitioning for Image Segmentation[][]
§ RandomWalks for Image Segmentation[][]
§ Blossom V:A new implementation of a minimum cost perfect matching algorithm[]
§ AnExperimental Comparison of Min-Cut/Max-Flow Algorithms for Energy Minimizationin Computer Vision[][]
§ GeodesicStar Convexity for Interactive Image Segmentation[]
§ ContourDetection and Image Segmentation Resources[][]
§ BiasedNormalized Cuts[]
§ Max-flow/min-cut[]
§ Chan-VeseSegmentation using Level Set[]
§ A Toolboxof Level Set Methods[]
§ Re-initializationFree Level Set Evolution via Reaction Diffusion[]
§ ImprovedC-V active contour model[][]
§ AVariational Multiphase Level Set Approach to Simultaneous Segmentation and BiasCorrection[][]
§ Level SetMethod Research by Chunming Li[]
§ ClassCutfor Unsupervised Class Segmentation[e]
§ SEEDS:Superpixels Extracted via Energy-Driven Sampling ][]
三、目标检测ObjectDetection：
§ A simpleobject detector with boosting []
§ INRIAObject Detection and Localization Toolkit [1] []
§ DiscriminativelyTrained Deformable Part Models [2] []
§ CascadeObject Detection with Deformable Part Models [3] []
§ Poselet[4] []
§ ImplicitShape Model [5] []
§ Viola andJones’s Face Detection [6] []
§ BayesianModelling of Dyanmic Scenes for Object Detection[][]
§ Handdetection using multiple proposals[]
§ ColorConstancy, Intrinsic Images, and Shape Estimation[][]
§ Discriminativelytrained deformable part models[]
§ GradientResponse Maps for Real-Time Detection of Texture-Less Objects: LineMOD []
§ ImageProcessing On Line[]
§ RobustOptical Flow Estimation[]
§ Where'sWaldo: Matching People in Images of Crowds[]
§ ScalableMulti-class Object Detection[]
§ Class-SpecificHough Forests for Object Detection[]
§ DeformedLattice Detection In Real-World Images[]
§ Discriminativelytrained deformable part models[]
四、显著性检测SaliencyDetection：
§ Itti,Koch, and Niebur’ saliency detection [1] []
§ Frequency-tunedsalient region detection [2] []
§ Saliencydetection using maximum symmetric surround [3] []
§ Attentionvia Information Maximization [4] []
§ Context-awaresaliency detection [5] []
§ Graph-basedvisual saliency [6] []
§ Saliencydetection: A spectral residual approach. [7] []
§ Segmentingsalient objects from images and videos. [8] []
§ SaliencyUsing Natural statistics. [9] []
§ DiscriminantSaliency for Visual Recognition from Cluttered Scenes. [10] []
§ Learningto Predict Where Humans Look [11] []
§ GlobalContrast based Salient Region Detection [12] []
§ BayesianSaliency via Low and Mid Level Cues[]
§ Top-DownVisual Saliency via Joint CRF and Dictionary Learning[][]
§ SaliencyDetection: A Spectral Residual Approach[]
五、图像分类、聚类ImageClassification, Clustering
§ PyramidMatch [1] []
§ SpatialPyramid Matching [2] []
§ Locality-constrainedLinear Coding [3] [] []
§ SparseCoding [4] [] []
§ TextureClassification [5] []
§ MultipleKernels for Image Classification [6] []
§ FeatureCombination [7] []
§ SuperParsing[]
§ LargeScale Correlation Clustering Optimization[]
§ Detectingand Sketching the Common[]
§ Self-TuningSpectral Clustering[][]
§ UserAssisted Separation of Reflections from a Single Image Using a Sparsity Prior[][]
§ Filtersfor Texture Classification[]
§ MultipleKernel Learning for Image Classification[]
§ SLICSuperpixels[]
六、抠图ImageMatting
§ A ClosedForm Solution to Natural Image Matting []
§ SpectralMatting []
§ Learning-basedMatting []
七、目标跟踪ObjectTracking：
§ A Forestof Sensors - Tracking Adaptive Background Mixture Models []
§ ObjectTracking via Partial Least Squares Analysis[][]
§ RobustObject Tracking with Online Multiple Instance Learning[][]
§ OnlineVisual Tracking with Histograms and Articulating Blocks[]
§ IncrementalLearning for Robust Visual Tracking[]
§ Real-timeCompressive Tracking[]
§ RobustObject Tracking via Sparsity-based Collaborative Model[]
§ VisualTracking via Adaptive Structural Local Sparse Appearance Model[]
§ OnlineDiscriminative Object Tracking with Local Sparse Representation[][]
§ SuperpixelTracking[]
§ LearningHierarchical Image Representation with Sparsity, Saliency and Locality[][]
§ OnlineMultiple Support Instance Tracking [][]
§ VisualTracking with Online Multiple Instance Learning[]
§ Objectdetection and recognition[]
§ CompressiveSensing Resources[]
§ RobustReal-Time Visual Tracking using Pixel-Wise Posteriors[]
§ Tracking-Learning-Detection[][]
§ the HandVu：vision-based hand gesture interface[]
§ LearningProbabilistic Non-Linear Latent Variable Models for Tracking Complex Activities[]
八、Kinect：
§ Kinecttoolbox[]
§ OpenNI[]
§ zouxy09CSDN Blog[]
§ FingerTracker手指跟踪[]
九、3D相关：
§ 3DReconstruction of a Moving Object[] []
§ Shape FromShading Using Linear Approximation[]
§ CombiningShape from Shading and Stereo Depth Maps[][]
§ Shape fromShading: A Survey[][]
§ ASpatio-Temporal Descriptor based on 3D Gradients (HOG3D)[][]
§ Multi-cameraScene Reconstruction via Graph Cuts[][]
§ A FastMarching Formulation of Perspective Shape from Shading under FrontalIllumination[][]
§ Reconstruction:3DShape, Illumination, Shading, Reflectance, Texture[]
§ MonocularTracking of 3D Human Motion with a Coordinated Mixture of Factor Analyzers[]
§ Learning3-D Scene Structure from a Single Still Image[]
十、机器学习算法：
§ Matlabclass for computing Approximate Nearest Nieghbor (ANN) [ providinginterface to]
§ RandomSampling[]
§ ProbabilisticLatent Semantic Analysis (pLSA)[]
§ FASTANNand FASTCLUSTER for approximate k-means (AKM)[]
§ FastIntersection / Additive Kernel SVMs[]
§ SVM[]
§ Ensemblelearning[]
§ DeepLearning[]
§ DeepLearning Methods for Vision[]
§ NeuralNetwork for Recognition of Handwritten Digits[]
§ Training adeep autoencoder or a classifier on MNIST digits[]
§ THE MNISTDATABASE of handwritten digits[]
§ Ersatz：deep neural networks in the cloud[]
§ DeepLearning []
§ sparseLM :Sparse Levenberg-Marquardt nonlinear least squares in C/C++[]
§ Weka 3:Data Mining Software in Java[]
§ Invitedtalk "A Tutorial on Deep Learning" by Dr. Kai Yu (余凯)[]
§ CNN -Convolutional neural network class[]
§ YannLeCun's Publications[]
§ LeNet-5,convolutional neural networks[]
§ Training adeep autoencoder or a classifier on MNIST digits[]
§ DeepLearning 大牛GeoffreyE. Hinton's HomePage[]
§ MultipleInstance Logistic Discriminant-based Metric Learning (MildML) and LogisticDiscriminant-based Metric Learning (LDML)[]
§ Sparsecoding simulation software[]
§ VisualRecognition and Machine Learning Summer School[]
十一、目标、行为识别Object,Action Recognition：
§ ActionRecognition by Dense Trajectories[][]
§ ActionRecognition Using a Distributed Representation of Pose and Appearance[]
§ RecognitionUsing Regions[][]
§ 2DArticulated Human Pose Estimation[]
§ Fast HumanPose Estimation Using Appearance and Motion via Multi-Dimensional BoostingRegression[][]
§ EstimatingHuman Pose from Occluded Images[][]
§ Quasi-densewide baseline matching[]
§ ChaLearnGesture Challenge: Principal motion: PCA-based reconstruction of motionhistograms[]
§ Real TimeHead Pose Estimation with Random Regression Forests[]
§ 2D ActionRecognition Serves 3D Human Pose Estimation[
§ A HoughTransform-Based Voting Framework for Action Recognition[
§ MotionInterchange Patterns for Action Recognition in Unconstrained Videos[
§ 2Darticulated human pose estimation software[]
§ Learningand detecting shape models []
§ ProgressiveSearch Space Reduction for Human Pose Estimation[]
§ LearningNon-Rigid 3D Shape from 2D Motion[]
十二、图像处理：
§ DistanceTransforms of Sampled Functions[]
§ TheComputer Vision Homepage[]
§ Efficientappearance distances between windows[]
§ ImageExploration algorithm[]
§ MotionMagnification 运动放大 []
§ BilateralFiltering for Gray and Color Images 双边滤波器 []
§ A FastApproximation of the Bilateral Filter using a Signal Processing Approach [
十三、一些实用工具：
§ EGT: aToolbox for Multiple View Geometry and Visual Servoing[] []
§ adevelopment kit of matlab mex functions for OpenCV library[]
§ FastArtificial Neural Network Library[]
十四、人手及指尖检测与识别：
§ finger-detection-and-gesture-recognition []
§ Hand andFinger Detection using JavaCV[]
§ Hand andfingers detection[]
十五、场景解释：
§ NonparametricScene Parsing via Label Transfer []
十六、光流Opticalflow：
§ Highaccuracy optical flow using a theory for warping []
§ DenseTrajectories Video Description []
§ SIFT Flow:Dense Correspondence across Scenes and its Applications[]
§ KLT: AnImplementation of the Kanade-Lucas-Tomasi Feature Tracker []
§ TrackingCars Using Optical Flow[]
§ Secrets ofoptical flow estimation and their principles[]
§ implmentationof the Black and Anandan dense optical flow method[]
§ OpticalFlow Computation[]
§ BeyondPixels: Exploring New Representations and Applications for Motion Analysis[]
§ A Databaseand Evaluation Methodology for Optical Flow[]
§ opticalflow relative[]
§ RobustOptical Flow Estimation []
§ opticalflow[]
十七、图像检索ImageRetrieval：
§ Semi-SupervisedDistance Metric Learning for Collaborative Image Retrieval ][]
十八、马尔科夫随机场MarkovRandom Fields：
§ MarkovRandom Fields for Super-Resolution ]
§ AComparative Study of Energy Minimization Methods for Markov Random Fields withSmoothness-Based Priors []
十九、运动检测Motiondetection：
§ MovingObject Extraction, Using Models or Analysis of Regions ]
§ BackgroundSubtraction: Experiments and Improvements for ViBe []
§ ASelf-Organizing Approach to Background Subtraction for Visual SurveillanceApplications []
§ changedetection.net:A new change detection benchmark dataset[]
§ ViBe - apowerful technique for background detection and subtraction in video sequences[]
§ BackgroundSubtraction Program[]
§ MotionDetection Algorithms[]
§ StuttgartArtificial Background Subtraction Dataset[]
§ ObjectDetection, Motion Estimation, and Tracking[]
Feature Detection and Description
General Libraries:
§ – Implementation of various featuredescriptors (including SIFT, HOG, and LBP) and covariant feature detectors(including DoG, Hessian, Harris Laplace, Hessian Laplace, Multiscale Hessian,Multiscale Harris). Easy-to-use Matlab interface. See – Slides providing a demonstration ofVLFeat and also links to other software. Check also
§ – Various implementations of modernfeature detectors and descriptors (SIFT, SURF, FAST, BRIEF, ORB, FREAK, etc.)
Fast Keypoint Detectors for Real-timeApplications:
§ – High-speed corner detectorimplementation for a wide variety of platforms
§ – Even faster than the FAST cornerdetector. A multi-scale version of this method is used for the BRISK descriptor(ECCV 2010).
Binary Descriptors for Real-TimeApplications:
§ – C++ code for a fast and accurateinterest point descriptor (not invariant to rotations and scale) (ECCV 2010)
§ – OpenCV implementation of theOriented-Brief (ORB) descriptor (invariant to rotations, but not scale)
§ – Efficient Binary descriptor invariantto rotations and scale. It includes a Matlab mex interface. (ICCV 2011)
§ – Faster than BRISK (invariant torotations and scale) (CVPR 2012)
SIFT and SURF Implementations:
§ SIFT: , , byDavid Lowe, ,
§ SURF: , ,
Other Local Feature Detectors andDescriptors:
§ – Oxford code for various affinecovariant feature detectors and descriptors.
§ – Source code for the Local Intensityorder Pattern (LIOP) descriptor (ICCV 2011).
§ – Source code for matching of local symmetryfeatures under large variations in lighting, age, and rendering style (CVPR2012).
Global Image Descriptors:
§ – Matlab code for the GIST descriptor
§ – Global visual descriptor for scenecategorization and object detection (PAMI 2011)
Feature Coding and Pooling
§ – Source code for variousstate-of-the-art feature encoding methods – including Standard hard encoding,Kernel codebook encoding, Locality-constrained linear encoding, and Fisherkernel encoding.
§ – Source code for feature pooling based on spatialpyramid matching (widely used for image classification)
Convolutional Nets and Deep Learning
§ – C++ Library for Energy-BasedLearning. It includes several demos and step-by-step instructions to trainclassifiers based on convolutional neural networks.
§ – Provides a matlab-like environmentfor state-of-the-art machine learning algorithms, including a fast implementationof convolutional neural networks.
§ -Various links for deep learning software.
Part-Based Models
§ – Library provided by the authors of the originalpaper (state-of-the-art in PASCAL VOC detection task)
§ – Branch-and-Bound implementation for adeformable part-based detector.
§ – Efficient implementation of a method thatachieves the exact same performance of deformable part-based detectors but withsignificant acceleration (ECCV 2012).
§ – Fast approach for deformable object detection(CVPR 2011).
§ – C++ and Matlab versions for objectdetection based on poselets.
§ – Implementation of a unified approachfor face detection, pose estimation, and landmark localization (CVPR 2012).
Attributes and Semantic Features
§ – Modified implementation of RankSVM totrain Relative Attributes (ICCV 2011).
§ – Implementation of object banksemantic features (NIPS 2010). See also
§ – Software for extracting high-levelimage descriptors (ECCV 2010, NIPS 2011, CVPR 2012).
Large-Scale Learning
§ – Source code for fast additive kernelSVM classifiers (PAMI 2013).
§ – Library for large-scale linear SVMclassification.
§ – Implementation for Pegasos SVM andHomogeneous Kernel map.
Fast Indexing and Image Retrieval
§ – Library for performing fastapproximate nearest neighbor.
§ – Source code for KernelizedLocality-Sensitive Hashing (ICCV 2009).
§ – Code for generation of small binary codes usingIterative Quantization and other baselines such as Locality-Sensitive-Hashing(CVPR 2011).
§ – Efficient code for state-of-the-art large-scaleimage retrieval (CVPR 2011).
Object Detection
§ See and above.
§ – Very fast and accurate pedestrian detector (CVPR2012).
§ – Excellent resource for pedestriandetection, with various links for state-of-the-art implementations.
§ – Enhanced implementation ofViola&Jones real-time object detector, with trained models for facedetection.
§ – Source code for branch-and-boundoptimization for efficient object localization (CVPR 2008).
3D Recognition
§ –Library for 3D image and point cloud processing.
Action Recognition
§ – Source code for action recognitionbased on the ActionBank representation (CVPR 2012).
§ –software for computing space-time interest point descriptors
§ – Look for Stacked ISA for Videos (CVPR 2011)
§ - C++ code for activity recognitionusing the velocity histories of tracked keypoints (ICCV 2009)
Datasets
Attributes
§ – 30,475 images of 50 animals classes with 6pre-extracted feature representations for each image.
§ – Attribute annotations for images collected fromYahoo and Pascal VOC 2008.
§ – 15,000 faces annotated with 10attributes and fiducial points.
§ – 58,797 face images of 200 people with73 attribute classifier outputs.
§ – 13,233 face images of 5,749 peoplewith 73 attribute classifier outputs.
§ – 8,000 people with annotatedattributes. Check also this for another dataset of humanattributes.
§ – Large-scale scene attribute database with ataxonomy of 102 attributes.
§ – Variety of attribute labels for theImageNet dataset.
§ – Data for OSR and a subset of PubFigdatasets. Check also this for the WhittleSearch data.
§ – Images of shopping categoriesassociated with textual descriptions.
Fine-grained Visual Categorization
§ – Hundreds of bird categories withannotated parts and attributes.
§ – 20,000 images of 120 breeds of dogsfrom around the world.
§ – 37 category pet dataset with roughly 200 imagesfor each class. Pixel level trimap segmentation is included.
§ – 832 images of 10 species ofbutterflies.
§ – Hundreds of flower categories.
Face Detection
§ – UMass face detection dataset andbenchmark (5,000+ faces)
§ – Classical face detection dataset.
Face Recognition
§ – Large collection of face recognition datasets.
§ – UMass unconstrained face recognitiondataset (13,000+ face images).
§ – includes face recognition grand challenge (FRGC),vendor tests (FRVT) and others.
§ –contains more than 750,000 images of 337 people, with 15 different views and 19lighting conditions.
§ – Classical face recognition dataset.
§ – Easy to use if you want play withsimple face datasets including Yale, ORL, PIE, and Extended Yale B.
§ – Low-resolution face dataset capturedfrom surveillance cameras.
Handwritten Digits
§ – large dataset containing a trainingset of 60,000 examples, and a test set of 10,000 examples.
Pedestrian Detection
§ – 10 hours of video taken from avehicle,350K bounding boxes for about 2.3K unique pedestrians.
§ – Currently one of the most popular pedestriandetection datasets.
§ – Urban dataset captured from a stereo rig mountedon a stroller.
§ – Dataset with image pairs recorded in an crowdedurban setting with an onboard camera.
§ – One of 20 categories in PASCAL VOCdetection challenges.
§ – Small dataset captured fromsurveillance cameras.
Generic Object Recognition
§ – Currently the largest visualrecognition dataset in terms of number of categories and images.
§ – 80 million 32x32 low resolutionimages.
§ – One of the most influential visualrecognition datasets.
§ / –Popular image datasets containing 101 and 256 object categories, respectively.
§ – Online annotation tool for buildingcomputer vision databases.
Scene Recognition
§ – MIT scene understanding dataset.
§ – Dataset of 15 natural scene categories.
Feature Detection and Description
§ – Widely used dataset for measuringperformance of feature detection and description. Checkfor an evaluation framework.
Action Recognition
§ – CVPR 2012 tutorial covering variousdatasets for action recognition.
RGBD Recognition
§ – Dataset containing 300 commonhousehold objects