This work presents Kornia - an open source computer vision library which consists of a set of differentiable routines and modules to solve generic computer vision problems. The package uses PyTorch as its main backend both for efficiency and to take advantage of the reverse-mode auto-differentiation to define and compute the gradient of complex functions. Inspired by OpenCV, Kornia is composed of a set of modules containing operators that can be inserted inside neural networks to train models to perform image transformations, camera calibration, epipolar geometry, and low level image processing techniques, such as filtering and edge detection that operate directly on high dimensional tensor representations. Examples of classical vision problems implemented using our framework are provided including a benchmark comparing to existing vision libraries.

Fine-grained categorization refers to the task of classifying objects that belong to the same basic-level class (e.g. different bird species) and share similar shape or visual appearances. Most of the state-of-the-art basic-level object classification algorithms have difficulties in this challenging problem. One reason for this can be attributed to the popular codebook-based image representation, often resulting in loss of subtle image information that are critical for fine-grained classification. Another way to address this problem is to introduce human annotations of object attributes or key points, a tedious process that is also difficult to generalize to new tasks. In this work, we propose a codebook-free and annotation-free approach for fine-grained image categorization. Instead of using vector-quantized codewords, we obtain an image representation by running a high throughput template matching process using a large number of randomly generated image templates. We then propose a novel bagging-based algorithm to build a final classifier by aggregating a set of discriminative yet largely uncorrelated classifiers. Experimental results show that our method outperforms state-of-the-art classification approaches on the Caltech-UCSD Birds dataset.

Feature matching is at the base of many computer vision problems, such as object recognition or structure from motion. Current methods rely on costly descriptors for detection and matching. In this paper, we propose a very fast binary descriptor based on BRIEF, called ORB, which is rotation invariant and resistant to noise. We demonstrate through experiments how ORB is at two orders of magnitude faster than SIFT, while performing as well in many situations. The efficiency is tested on several real-world applications, including object detection and patch-tracking on a smart phone.

A robust robot perception system intended to enable object manipulation needs to be able to accurately identify objects and their pose at high speeds. Since objects vary considerably in surface properties, rigidity and articulation, no single detector or object estimation method has been shown to provide reliable detection across object types to date. This indicates the need for an architecture that is able to quickly swap detectors, pose estimators, and filters, or to run them in parallel or serial and combine their results, preferably without any code modifications at all. In this paper, we present our implementation of such an infrastructure, ReIn (REcognition INfrastructure), to answer these needs. ReIn is able to combine a multitude of 2D/3D object recognition and pose estimation techniques in parallel as dynamically loadable plugins. It also provides an extremely efficient data passing architecture, and offers the possibility to change the parameters and initial settings of these techniques during their execution. In the course of this work we introduce two new classifiers designed for robot perception needs: BiGGPy (Binarized Gradient Grid Pyramids) for scalable 2D classification and VFH (Viewpoint Feature Histograms) for 3D classification and pose. We then show how these two classifiers can be easily combined using ReIn to solve object recognition and pose identification problems.

We describe an autonomous robotic system capable of navigating through an office environment, opening doors along the way, and plugging itself into electrical outlets to recharge as needed. We demonstrate through extensive experimentation that our robot executes these tasks reliably, without requiring any modification to the environment. We present robust detection algorithms for doors, door handles, and electrical plugs and sockets, combining vision and laser sensors. We show how to overcome the unavoidable shortcoming of perception by integrating compliant control into manipulation motions. We present a visual-differencing approach to high-precision plug-insertion that avoids the need for high-precision hand-eye calibration.

We present the Viewpoint Feature Histogram (VFH), a descriptor for 3D point cloud data that encodes geometry and viewpoint. We demonstrate experimentally on a set of 60 objects captured with stereo cameras that VFH can be used as a distinctive signature, allowing simultaneous recognition of the object and its pose. The pose is accurate enough for robot manipulation, and the computational cost is low enough for real time operation. VFH was designed to be robust to large surface noise and missing depth information in order to work reliably on stereo data.

Existing methods for recognition of object instances and categories based on quantized local features can perform poorly when local features exist on transparent surfaces, such as glass or plastic objects. There are characteristic patterns to the local appearance of transparent objects, but they may not be well captured by distances to individual examples or by a local pattern codebook obtained by vector quantization. The appearance of a transparent patch is determined in part by the refraction of a background pattern through a transparent medium: the energy from the background usually dominates the patch appearance. We model transparent local patch appearance using an additive model of latent factors: background factors due to scene content, and factors which capture a local edge energy distribution characteristic of the refraction. We implement our method using a novel LDA-SIFT formulation which performs LDA prior to any vector quantization step; we discover latent topics which are characteristic of particular transparent patches and quantize the SIFT space into transparent visual words according to the latent topic dimensions. No knowledge of the background scene is required at test time; we show examples recognizing transparent glasses in a domestic environment.

This paper evaluates the suitability of the MapReduce model for multi-core and multi-processor systems. MapReduce was created by Google for application development on data-centers with thousands of servers. It allows programmers to write functional-style code that is automatically parallelized and scheduled in a distributed system. We describe Phoenix, an implementation of MapReduce for shared-memory systems that includes a programming API and an efficient runtime system. The Phoenix runtime automatically manages thread creation, dynamic task scheduling, data partitioning, and fault tolerance across processor nodes. We study Phoenix with multi-core and symmetric multiprocessor systems and evaluate its performance potential and error recovery features. We also compare MapReduce code to code written in lower-level APIs such as P-threads. Overall, we establish that, given a careful implementation, MapReduce is a promising model for scalable performance on shared-memory systems with simple parallel code

This paper describes a hierarchical Bayesian network used for segmenting desert images and detecting off road drivable corridors for autonomous navigation. Unlike the embedded hidden Markov model the Bayesian network presented in this paper can successfully account for natural dependencies between neighboring pixels in both image dimensions making it more suitable for a larger class of images. The method described here was developed within the Stanford racing team that won the DARPA Grand Challenge 2005 after driving over 130 miles autonomously in the Nevada desert.

We are at the beginning of the multicore era. Computers will have increasingly many cores (processors), but there is still no good programming framework for these architectures, and thus no simple and unified way for machine learning to take advantage of the potential speed up. In this paper, we develop a broadly applicable parallel programming method, one that is easily applied to many different learning algorithms. Our work is in distinct contrast to the tradition in machine learning of designing (often ingenious) ways to speed up a single algorithm at a time. Specifically, we show that algorithms that fit the Statistical Query model [15] can be written in a certain "summation form," which allows them to be easily parallelized on multicore computers. We adapt Google's map-reduce [7] paradigm to demonstrate this parallel speed up technique on a variety of learning algorithms including locally weighted linear regression (LWLR), k-means, logistic regression (LR), naive Bayes (NB), SVM, ICA, PCA, gaussian discriminant analysis (GDA), EM, and backpropagation (NN). Our experimental results show basically linear speedup with an increasing number of processors.

This paper presents a new filter for online data association problems in high-dimensional spaces. The key innovation is a representation of the data association posterior in information form, in which the “proxim- ity” of objects and tracks are expressed by numerical links. Updating these links requires linear time, compared to exponential time required for computing the exact posterior probabilities. The paper derives the algorithm formally and provides comparative results using data obtained by a real-world camera array and by a large-scale sensor network simu- lation.

A method for a real-time vision system that automatically detects a user's eye blinks and accurately measures their durations is introduced. The system is intended to provide an alternate input modality to allow people with severe disabilities to access a computer. Voluntary long blinks trigger mouse clicks, while involuntary short blinks are ignored. The system enables communication using "blink patterns:" sequences of long and short blinks which are interpreted as semiotic messages. The location of the eyes is determined automatically through the motion of the user's initial blinks. Subsequently, the eye is tracked by correlation across time, and appearance changes are automatically analyzed in order to classify the eye as either open or closed at each frame. No manual initialization, special lighting, or prior face detection is required. The system has been tested with interactive games and a spelling program. Results demonstrate overall detection accuracy of 95.61% and an average rate of 28 frames per second.

Intel's Microcomputer Research Lab has been developing a highly optimized Computer Vision Library (CVLib) that automatically detects processor type and loads the appropriate MMX/sup TM/ technology assembly tuned module for that processor. MMX optimized functions are from 2 to 8 times faster than optimized C functions. We will be demonstrating various algorithms supported by CVLib and handing out CDs containing the library.

This paper describes a stereo-based approach for gesture recognition that works well under extreme lighting conditions and tolerates a large range of hand poses. The approach proposes a hybrid gesture representation that models the user's arm as a 3D line and uses images to represent the hand gestures. The algorithm finds the arm orientation and the hand location from the disparity data and uses this information to initialize a color based segmentation algorithm that cleanly separates the hand from the background. Finally, our approach uses the arm orientation to compute a frontal view of the hand through perspective unwarping producing easily recognizable hand gesture templates. The classification algorithm uses statistical moments of the binarized gesture templates to find the match. We achieve 96% recognition rates under varying lighting and 3D poses for a set of 6 gestures.

This paper uses a simple method for representing motion in successively layered silhouettes that directly encode system time termed the timed Motion History Image (tMHI). This representation can be used to both (a) determine the current pose of the object and to (b) segment and measure the motions induced by the object in a video scene. These segmented regions are not "motion blobs", but instead motion regions naturally connected to the moving parts of the object of interest. This method may be used as a very general gesture recognition "toolbox". We use it to recognize waving and overhead clapping motions to control a music synthesis program.

As a step towards a perceptual user interface, an object tracking algorithm is developed and demonstrated tracking human faces. Computer vision algorithms that are intended to form part of a perceptual user interface must be fast and efficient. They must be able to track in real time and yet not absorb a major share of computational resources. An efficient, new algorithm is described here based on the mean shift algorithm. The mean shift algorithm robustly finds the mode (peak) of probability distributions. We first describe histogram based methods of producing object probability distributions. In our case, we want to track the mode of an object's probability distribution within a video scene. Since the probability distribution of the object can change and move dynamically in time, the mean shift algorithm is modified to deal with dynamically changing probability distributions. The modified algorithm is called the Continuously Adaptive Mean Shift (CAMSHIFT) algorithm. CAMSHIFT is then used as an interface for games and graphics.

The recognition of three-dimensional (3-D) objects from sequences of their two-dimensional (2-D) views is modeled by a family of self-organizing neural architectures, called VIEWNET, that use View Information Encoded With NETworks. VIEWNET incorporates a preprocessor that generates a compressed but 2-D invariant representation of an image, a supervised incremental learning system that classifies the preprocessed representations into 2-D view categories whose outputs are combined into 3-D invariant object categories, and a working memory that makes a 3-D object prediction by accumulating evidence from 3-D object category nodes as multiple 2-D views are experienced. The simplest VIEWNET achieves high recognition scores without the need to explicitly code the temporal order of 2-D views in working memory. Working memories are also discussed that save memory resources by implicitly coding temporal order in terms of the relative activity of 2-D view category nodes, rather than as explicit 2-D view transitions. Variants of the VIEWNET architecture may be used for scene understanding by using a preprocessor and classier that can determine both what objects are in a scene and where they are located. The present VIEWNET preprocessor includes the CORT-X 2 filter, which discounts the illuminant, regularizes and completes figural boundaries, and suppresses image noise. This boundary segmentation is rendered invariant under 2-D translation, rotation, and dilation by use of a log polar transform. The invariant spectra undergo Gaussian coarse coding to further reduce noise and 3-D foreshortening effects, and to increase generalization. These compressed codes are input into the classifier, a supervised learning system based on the fuzzy ARTMAP algorithm. Fuzzy ARTMAP learns 2-D view categories that are invariant under 2-D image translation, rotation, and dilation as well as 3-D image transformations that do not cause a predictive error. Evidence from sequences of 2-D view categories converges at 3-D object nodes that generate a response invariant under changes of 2-D view. These 3-D object nodes input to a working memory that accumulates evidence over time to improve object recognition. In the simplest working memory, each occurrence (nonoccurrence) of a 2-D view category increases (decreases) the corresponding node's activity in working memory. The maximally active node is used to predict the 3-D object. Recognition is studied with noisy and clean images using slow and fast learning. Slow learning at thefuzzy ARTMAP mapfield is adapted to learn the conditional probability of the 3-D object given the selected 2-D view category. VIEWNET is demonstrated on an MIT Lincoln Laboratory database of 128 × 128 2-D views of aircraft with and without additive noise. A recognition rate of up to 90% is achieved with one 2-D view and of up to 98.5% correct with three 2-D views. The properties of 2-D view and 3-D object category nodes are compared with those of cells in monkey inferotemporal cortex.

Working memory neural networks, called Sustained Temporal Order REcurrent (STORE) models, encode the invariant temporal order of sequential events in short-term memory (STM). Inputs to the networks may be presented with widely differing growth rates, amplitudes, durations, and interstimulus intervals without altering the stored STM representation. The STORE temporal order code is designed to enable groupings of the stored events to be stably learned and remembered in real time, even as new events perturb the system. Such invariance and stability properties are needed in neural architectures which self-organize learned codes for variable-rate speech perception, sensorimotor planning, or three-dimensional (3-D) visual object recognition. Using such a working memory, a self-organizing architecture for invariant 3-D visual object recognition is described. The new model is based on the model of Seibert and Waxman (1990a), which builds a 3-D representation of an object from a temporally ordered sequence of its two-dimensional (2-D) aspect graphs. The new model, called an ARTSTORE model, consists of the following cascade of processing modules: Invariant Preprocessor → ART 2 → STORE Model → ART 2 → Outstar Network.