Visual Recognition and Learning

Visual recognition is the process of automatically extracting valuable information from visual data (images and videos). This includes: knowing what objects are in a picture, where they are, what's happening, and where it is happening. Given the nature of the problem and the abundanc e of data available today, visual recognition is typically formulated as a machine learning problem.

In this course we study the problems of instance-level object recognition ("what is this?"), category-level object recognition ("what kind of thing is this?"), attributes ("what is this thing like?"), parts ("what is this thing composed of?"), action recognition ("what's going on ?"), pose estimation ("what is the configuration of this thing?"), detection ("where is this thing in the image?"), and segmentation ("where e xactly is this thing in the image?"). We will cover a wide range of supervised, semi-supervised, and unsupervised models (consisting of fully- , partially-, and un-labeled data), as well as transferable models in object and activity recognition (learning from one dataset or task and applying to a different one).

Students are recommended to be familiar with either computer vision or machine learning. If you are interested in this course and you have been exposed to neither computer vision nor machine learning please ask for instructor's approval. Throughout the quarter I will provide necessary background material.

This course will consist of lectures, student presentations, and a course project. Students who register for only one credit are not required to do a course project.

• Background Material

• Introduction[1.pptx]
• Features and Representations[2.pptx][3.pptx]
• Standard Tools and Techniques[4.pptx][5.pptx][6.pptx] [7.pptx][8.pptx][9.pptx]

• Object Recognition

• Object Detection
• Main: A Discriminatively Trained, Multiscale, Deformable Part Model
• Main: Object Detection with Discriminatively Trained Part Based Models
• Speedup: Cascade Object Detection with Deformable Part Models
• Analysis: How important are Deformable Parts in the Deformable Parts Model?
• Analysis: Diagnosing Error in Object Detectors
• Pictorial Structures: Pictorial structures for object recognition
• Latent SVM: Support vector machines for multiple-instance learning
• Object Instance Recognition and Matching [Ji]
• Video Google: A Text Retrieval Approach to Object Matching in Videos
• Large Scale: Discovering Favorite Views of Popular Places with Iconoid Shift
• Image Webs: Computing and Exploiting Connectivity in Image Collections
• Object Retrieval with Large Vocabularies and Fast Spatial Matching
• City-Scale Location Recognition
• Object Category Recognition [Supasorn]
• Locality-Constrained Linear Coding for Image Classification
• Beyond Sliding Windows: Object Localization by Efficient Subwindow Search
• Visual Categorization with Bags of Keypoints
• Role of Context
• Using the Forest to See the Trees: Exploiting Context for Visual Object Detection and Localization
• Contextual Priming for Object Detection
• Discriminative Models for Multi-Class Object Layout
• An Empirical Study of Context in Object Detection
• Stuff, Things, Geometry and Scenes [Ezgi,Andre]
• Learning Spatial Context: Using Stuff to Find Things
• Blocks World Revisited: Image Understanding Using Qualitative Geometry and Mechanics
• Thinking Inside the Box: Using Appearance Models and Context Based on Room Geometry
• Scene Semantics from Long-term Observation of People
• Context Based Object Categorization: A Critical Survey
• Weakly supervised Recognition [Yuyin]
• Self-paced Learning for Latent Variable Models
• Measuring the objectness of image windows
• Unsupervised Recognition and Object Discovery [Tanner]
• Discovering objects and their location in images
• Large Scale Recognition [Jinna]
• Total Recall II: Query Expansion Revisited
• Hedging Your Bets: Optimizing Accuracy-Specificity Trade-offs in Large Scale Visual Recognition
• What Does Classifying More than 10,000 Image Categories Tell Us?
• Comparative Object Similarity for Improved Recognition with Few or Zero Examples
• Parts and Attributes [Ila]
• Describing Objects by Their Attributes
• Relative Attributes
• Poselets: Body Part Detectors Trained Using 3D Human Pose Annotations
• Shared Representations, Transfer Learning
• Sharing Visual Features for Multiclass and Multiview Object Detection
• Tabula Rasa: Model Transfer for Object Category Detection
• Learning to Recognize Activities from the Wrong View Point
• Deep Architectures [Rob]
• Discriminative Learning of Sum-Product Networks
• RGB-(D) Scene Labeling: Features and Algorithms

• Action Recognition

• From Video [Yongjin]
• Evaluation of local spatio-temporal features for action recognition
• Learning realistic human actions from movies
• Matching Local Self-Similarities across Images and Videos
• View-Independent Action Recognition from Temporal Self-Similarities
• From Still Images [Alex]
• Action Recognition from a Distributed Representation of Pose and Appearance
• Observing Human-Object Interactions: Using Spatial and Functional Compatibility for Recognition
• Pose Estimation [Youngdae]
• Real-Time Human Pose Recognition in Parts from a Single Depth Image
• Cascaded Models for Articulated Pose Estimation
• Articulated pose estimation with flexible mixtures-of-parts
• Objects, Actions, and Poses [Jun]
• Detecting Actions, Poses, and Objects with Relational Phraselets
• Objects in Action: An Approach for Combining Action Understanding and Object Perception
• Modeling Mutual Context of Object and Human Pose in Human-Object Interaction Activities

• Segmentation [Peter]

• Category Independent Object Proposals
• Constrained Parametric Min-Cuts for Automatic Object Segmentation
• Shape Sharing for Object Segmentation