Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day.

Gradient-based learning applied to document recognition

We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

There is, I think, something ethereal about i —the square root of minus one. I remember first hearing about it at school. It seemed an odd beast at that time—an intruder hovering on the edge of reality.

Usually familiarity dulls this sense of the bizarre, but in the case of i it was the reverse: over the years the sense of its surreal nature intensified. It seemed that it was impossible to write mathematics that described the real world in …

I and i

We study the question of feature sets for robust visual object recognition; adopting linear SVM based human detection as a test case. After reviewing existing edge and gradient based descriptors, we show experimentally that grids of histograms of oriented gradient (HOG) descriptors significantly outperform existing feature sets for human detection. We study the influence of each stage of the computation on performance, concluding that fine-scale gradients, fine orientation binning, relatively coarse spatial binning, and high-quality local contrast normalization in overlapping descriptor blocks are all important for good results. The new approach gives near-perfect separation on the original MIT pedestrian database, so we introduce a more challenging dataset containing over 1800 annotated human images with a large range of pose variations and backgrounds.

/pdf/histograms-of-oriented-gradients-for-human-detection-3nteinziuq.pdf

Histograms of oriented gradients for human detection

Object detection performance, as measured on the canonical PASCAL VOC dataset, has plateaued in the last few years. The best-performing methods are complex ensemble systems that typically combine multiple low-level image features with high-level context. In this paper, we propose a simple and scalable detection algorithm that improves mean average precision (mAP) by more than 30% relative to the previous best result on VOC 2012 -- achieving a mAP of 53.3%. Our approach combines two key insights: (1) one can apply high-capacity convolutional neural networks (CNNs) to bottom-up region proposals in order to localize and segment objects and (2) when labeled training data is scarce, supervised pre-training for an auxiliary task, followed by domain-specific fine-tuning, yields a significant performance boost. Since we combine region proposals with CNNs, we call our method R-CNN: Regions with CNN features. We also present experiments that provide insight into what the network learns, revealing a rich hierarchy of image features. Source code for the complete system is available at http://www.cs.berkeley.edu/~rbg/rcnn.

/pdf/rich-feature-hierarchies-for-accurate-object-detection-and-22fvjlggyg.pdf

Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation

We introduce the concept of a Visual Compiler that generates a scene specific pedestrian detector and pose estimator without any pedestrian observations. Given a single image and auxiliary scene information in the form of camera parameters and geometric layout of the scene, the Visual Compiler first infers geometrically and photometrically accurate images of humans in that scene through the use of computer graphics rendering. Using these renders we learn a scene-and-region specific spatially-varying fully convolutional neural network, for simultaneous detection, pose estimation and segmentation of pedestrians. We demonstrate that when real human annotated data is scarce or non-existent, our data generation strategy can provide an excellent solution for bootstrapping human detection and pose estimation. Experimental results show that our approach outperforms off-the-shelf state-of-the-art pedestrian detectors and pose estimators that are trained on real data.

Visual Compiler: Synthesizing a Scene-Specific Pedestrian Detector and Pose Estimator

Recording dynamic events, such as a sports event, a ballet performance, or a lecture, digitally for experiencing in a spatiotemporally distant setting requires 4D capture: three dimensions for their geometry/ appearance over the fourth dimension of time. Cameras are suitable for this task as they are nonintrusive, universal, and inexpensive. Computer Vision techniques have advanced sufficiently to make the 4D capture possible. In this paper, we present a historical perspective on the Virtualized RealityTM system developed since early 90s to early 2000 at CMU for the 4D capture of dynamic events.

/pdf/historical-perspectives-on-4d-virtualized-reality-414gmk56lp.pdf

Historical Perspectives on 4D Virtualized Reality

This paper presents a model-based object recognition method which combines a bottom-up evidence accumulation process and a top-down hypothesis verification process. The hypothesize-and-test paradigm is fundamental in model-based vision. However, research issues remain on how the bottom-up process gathers pieces of evidence and when the top-down process should take the lead. To accumulate pieces of evidence, we use a configuration space whose points represent a configuration of an object (ie. position and orientation of an object in an image). If a feature is found which matches a part of an object model, the configuration space is updated to reflect the possible configurations of the object. A region in the configuration space where multiple pieces of evidence from such feature-part matches overlap suggests a high probability that the object exists in an image with a configuration in that region. The cost of the bottom-up process to further accumulate evidence for localization, and that of the top-down process to recognize the object by verification, are compared by considering the size of the search region and the probability of success of verification. If the cost of the top-down process becomes lower, hypotheses are generated and their verification processes are started. The first version of the recognition program has been written and applied to the recognition of a jet airplane in synthetic aperture radar (SAR) images. In creating a model of an object, we have used a SAR simulator as a sensor model, so that we can predict those object features which are reliably detectable by the sensors. The program is being tested with simulated SAR images, and shows promising performance.

Model-Based Vision By Cooperative Processing Of Evidence And Hypotheses Using Configuration Spaces

Image understanding research at Carnegie Mellon

The authors present a method for determining the shapes of surfaces whose reflectance properties may vary from Lambertian to specular, without prior knowledge of the relative strengths of the Lambertian and specular components of reflection. The object surface is illuminated using extended light sources and is viewed from a single direction. Surface illumination using extended sources makes it possible to ensure the detection of both Lambertian and specular reflections. Multiple source directions are used to obtain an image sequence of the object. An extraction algorithm uses the set of image intensity values measured at each surface point to compute orientation as well as relative strengths of the Lambertian and specular reflection components. The proposed method is called photometric sampling, as it uses samples of photometric function that relates image intensity to surface orientation, reflectance, and light source characteristics. Experiments were conducted on Lambertian surfaces, specular surfaces, and hybrid surfaces, whose reflectance models are composed of both Lambertian and specular components. The results show high accuracy in measured orientations and estimated reflectance parameters. >

Takeo Kanade

Papers

Visual Compiler: Synthesizing a Scene-Specific Pedestrian Detector and Pose Estimator

Historical Perspectives on 4D Virtualized Reality

Model-Based Vision By Cooperative Processing Of Evidence And Hypotheses Using Configuration Spaces

Image understanding research at Carnegie Mellon

Shape and reflectance from an image sequence generated using extended sources