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

We present YOLO, a new approach to object detection. Prior work on object detection repurposes classifiers to perform detection. Instead, we frame object detection as a regression problem to spatially separated bounding boxes and associated class probabilities. A single neural network predicts bounding boxes and class probabilities directly from full images in one evaluation. Since the whole detection pipeline is a single network, it can be optimized end-to-end directly on detection performance. Our unified architecture is extremely fast. Our base YOLO model processes images in real-time at 45 frames per second. A smaller version of the network, Fast YOLO, processes an astounding 155 frames per second while still achieving double the mAP of other real-time detectors. Compared to state-of-the-art detection systems, YOLO makes more localization errors but is less likely to predict false positives on background. Finally, YOLO learns very general representations of objects. It outperforms other detection methods, including DPM and R-CNN, when generalizing from natural images to other domains like artwork.

/pdf/you-only-look-once-unified-real-time-object-detection-3gan6roe1o.pdf

You Only Look Once: Unified, Real-Time Object Detection

This paper describes a machine learning approach for visual object detection which is capable of processing images extremely rapidly and achieving high detection rates. This work is distinguished by three key contributions. The first is the introduction of a new image representation called the "integral image" which allows the features used by our detector to be computed very quickly. The second is a learning algorithm, based on AdaBoost, which selects a small number of critical visual features from a larger set and yields extremely efficient classifiers. The third contribution is a method for combining increasingly more complex classifiers in a "cascade" which allows background regions of the image to be quickly discarded while spending more computation on promising object-like regions. The cascade can be viewed as an object specific focus-of-attention mechanism which unlike previous approaches provides statistical guarantees that discarded regions are unlikely to contain the object of interest. In the domain of face detection the system yields detection rates comparable to the best previous systems. Used in real-time applications, the detector runs at 15 frames per second without resorting to image differencing or skin color detection.

/pdf/rapid-object-detection-using-a-boosted-cascade-of-simple-4crej3dxmd.pdf

Rapid object detection using a boosted cascade of simple features

The Pascal Visual Object Classes (VOC) challenge is a benchmark in visual object category recognition and detection, providing the vision and machine learning communities with a standard dataset of images and annotation, and standard evaluation procedures. Organised annually from 2005 to present, the challenge and its associated dataset has become accepted as the benchmark for object detection.

This paper describes the dataset and evaluation procedure. We review the state-of-the-art in evaluated methods for both classification and detection, analyse whether the methods are statistically different, what they are learning from the images (e.g. the object or its context), and what the methods find easy or confuse. The paper concludes with lessons learnt in the three year history of the challenge, and proposes directions for future improvement and extension.

/pdf/the-pascal-visual-object-classes-voc-challenge-ugvl0p2h51.pdf

The Pascal Visual Object Classes (VOC) Challenge

This paper proposes a Fast Region-based Convolutional Network method (Fast R-CNN) for object detection. Fast R-CNN builds on previous work to efficiently classify object proposals using deep convolutional networks. Compared to previous work, Fast R-CNN employs several innovations to improve training and testing speed while also increasing detection accuracy. Fast R-CNN trains the very deep VGG16 network 9x faster than R-CNN, is 213x faster at test-time, and achieves a higher mAP on PASCAL VOC 2012. Compared to SPPnet, Fast R-CNN trains VGG16 3x faster, tests 10x faster, and is more accurate. Fast R-CNN is implemented in Python and C++ (using Caffe) and is available under the open-source MIT License at https://github.com/rbgirshick/fast-rcnn.

Fast R-CNN

Annotation recognition and parsing is accomplished by first recognizing and grouping shapes such that relationships between the annotations and the underlying text and/or images can be determined. The recognition and grouping is followed by categorization of recognized annotations according to predefined types. The classification may be according to functionality, relation to content, and the like. In a third phase, the annotations are anchored to the underlying text or images they are found to be related to.

Parsing of ink annotations

Document features or document ranking values can be associated with a distribution of values. Feature values, feature value coefficients, and/or document ranking values can be generated based on sampled values from the distribution of values. This can allow the relative ranking of a document to vary. As additional information is obtained regarding the document, leading to greater certainty about the appropriate ranking of the document, the width or variation generated by the distribution can be reduced to provide more stable ranking values

Ranking search results using feature score distributions

Image feature selection and extraction (e.g., for image classifier training) is accomplished in an integrated manner, such that higher-order features are merely developed from first-order features selected for image classification. That is, first-order image features are selected for image classification from an image feature pool, initially populated with pre-extracted first-order image features. The selected first-order classifying features are paired with previously selected first-order classifying features to generate higher-order features. The higher-order features are placed into the image feature pool as they are developed or “on-the-fly” (e.g., for use in image classifier training).

Feature selection and extraction

Image recognition is utilized to facilitate in scoring parse trees for two-dimensional recognition tasks. Trees and subtrees are rendered as images and then utilized to determine parsing scores. Other instances of the subject invention can incorporate additional features such as stroke curvature and/or nearby white space as rendered images as well. Geometric constraints can also be employed to increase performance of a parsing process, substantially improving parsing speed, some even resolvable in polynomial time. Additional performance enhancements can be achieved in yet other instances of the subject invention by employing constellations of integral images and/or integral images of document features.

Application of grammatical parsing to visual recognition tasks

Systems and methods are described for face recognition using discriminatively trained orthogonal rank one tensor projections. In an exemplary system, images are treated as tensors, rather than as conventional vectors of pixels. During runtime, the system designs visual features—embodied as tensor projections—that minimize intraclass differences between instances of the same face while maximizing interclass differences between the face and faces of different people. Tensor projections are pursued sequentially over a training set of images and take the form of a rank one tensor, i.e., the outer product of a set of vectors. An exemplary technique ensures that the tensor projections are orthogonal to one another, thereby increasing ability to generalize and discriminate image features over conventional techniques. Orthogonality among tensor projections is maintained by iteratively solving an ortho-constrained eigenvalue problem in one dimension of a tensor while solving unconstrained eigenvalue problems in additional dimensions of the tensor.

Paul A. Viola

Papers

Parsing of ink annotations

Ranking search results using feature score distributions

Feature selection and extraction

Application of grammatical parsing to visual recognition tasks

Face recognition using discriminatively trained orthogonal tensor projections