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

In this paper we describe a trainable object detector and its instantiations for detecting faces and cars at any size, location, and pose. To cope with variation in object orientation, the detector uses multiple classifiers, each spanning a different range of orientation. Each of these classifiers determines whether the object is present at a specified size within a fixed-size image window. To find the object at any location and size, these classifiers scan the image exhaustively.

Each classifier is based on the statistics of localized parts. Each part is a transform from a subset of wavelet coefficients to a discrete set of values. Such parts are designed to capture various combinations of locality in space, frequency, and orientation. In building each classifier, we gathered the class-conditional statistics of these part values from representative samples of object and non-object images. We trained each classifier to minimize classification error on the training set by using Adaboost with Confidence-Weighted Predictions (Shapire and Singer, 1999). In detection, each classifier computes the part values within the image window and looks up their associated class-conditional probabilities. The classifier then makes a decision by applying a likelihood ratio test. For efficiency, the classifier evaluates this likelihood ratio in stages. At each stage, the classifier compares the partial likelihood ratio to a threshold and makes a decision about whether to cease evaluation—labeling the input as non-object—or to continue further evaluation. The detector orders these stages of evaluation from a low-resolution to a high-resolution search of the image. Our trainable object detector achieves reliable and efficient detection of human faces and passenger cars with out-of-plane rotation.

/pdf/object-detection-using-the-statistics-of-parts-23wiynfqr5.pdf

Object Detection Using the Statistics of Parts

There has been little clinical research to examine the effects of patient positioning and pelvic motion on the alignment of the acetabular implant during total hip replacement surgery. Until now, no tools were capable of accurately measuring these variables during the actual procedure. As part of a broader program in medical robotics and computer assisted surgery, a clinical system has been developed that includes several enabling technologies. The hip navigation system (HipNav) continuously and precisely measures pelvic location and tracks relative implant alignment intraoperatively. HipNav technology is used to gauge current clinical practice and provide intraoperative feedback to surgeons with the goal of improving the precision and accuracy of acetabular alignment during total hip replacement. This system provides surgeons with a new class of image guided measurement tools and assist devices. These tools successfully were introduced into the clinical practice of surgery with results showing the following: (1) There exist unpredictable and large variations in the initial position of patients' pelves on the operating room table and significant pelvic movement during surgery and during intraoperative range of motion testing; (2) current mechanical acetabular alignment guides do not account for these variations, and result in variable and in the majority of cases unacceptable acetabular alignment; and (3) press fitting oversized acetabular components influences the final cup orientation.

The Otto Aufranc Award. Image guided navigation system to measure intraoperatively acetabular implant alignment.

Digital video is rapidly becoming important for education, entertainment, and a host of multimedia applications. With the size of the video collections growing to thousands of hours, technology is needed to effectively browse segments in a short time without losing the content of the video. We propose a method to extract the significant audio and video information and create a "skim" video which represents a very short synopsis of the original. The goal of this work is to show the utility of integrating language and image understanding techniques for video skimming by extraction of significant information, such as specific objects, audio keywords and relevant video structure. The resulting skim video is much shorter, where compaction is as high as 20:1, and yet retains the essential content of the original segment.

Video skimming and characterization through the combination of image and language understanding techniques

This paper presents algorithms for identifying parameters of multi-degrees-of-freedom robotic arm. First, we outline the fundamental properties of the Newton-Euler formulation of robot dynamics from the view point of parameter identification. We then show that the Newton-Euler model which is nonlinear in some of dynamic parameters can be transformed into the equivalent model which is linear in dynamic parameters. We develop both on-line and off-line parameter estimation procedures. To illustrate our approach, we identify the dynamic parameters of the cylindrical robot, and the three degree-of-freedom positioning system of the CMU Direct-Drive Arm II.

Parameter identification of robot dynamics

Reconstructing an arbitrary configuration of 3D points from their projection in an image is an ill-posed problem. When the points hold semantic meaning, such as anatomical landmarks on a body, human observers can often infer a plausible 3D configuration, drawing on extensive visual memory. We present an activity-independent method to recover the 3D configuration of a human figure from 2D locations of anatomical landmarks in a single image, leveraging a large motion capture corpus as a proxy for visual memory. Our method solves for anthropometrically regular body pose and explicitly estimates the camera via a matching pursuit algorithm operating on the image projections. Anthropometric regularity (i.e., that limbs obey known proportions) is a highly informative prior, but directly applying such constraints is intractable. Instead, we enforce a necessary condition on the sum of squared limb-lengths that can be solved for in closed form to discourage implausible configurations in 3D. We evaluate performance on a wide variety of human poses captured from different viewpoints and show generalization to novel 3D configurations and robustness to missing data.

/pdf/reconstructing-3d-human-pose-from-2d-image-landmarks-b5g9lqff8i.pdf

Takeo Kanade

Papers

Object Detection Using the Statistics of Parts

The Otto Aufranc Award. Image guided navigation system to measure intraoperatively acetabular implant alignment.

Video skimming and characterization through the combination of image and language understanding techniques

Parameter identification of robot dynamics

Reconstructing 3d human pose from 2d image landmarks