Abstract: Wireless capsule endoscopy involves inspection of hours of video material by a highly qualified professional. Time episodes corresponding to intestinal contractions, which are of interest to the physician constitute about 1% of the video. The problem is to label automatically time episodes containing contractions so that only a fraction of the video needs inspection. As the classes of contraction and non-contraction images in the video are largely imbalanced, ROC curves are used to optimize the trade-off between false positive and false negative rates. Classifier ensemble methods and simple classifiers were examined. Our results reinforce the claims from recent literature that classifier ensemble methods specifically designed for imbalanced problems have substantial advantages over simple classifiers and standard classifier ensembles. By using ROC curves with the bagging ensemble method the inspection time can be drastically reduced at the expense of a small fraction of missed contractions.

Abstract: IF: 0.552

Abstract: In this paper we propose a novel approach for writer adaptation in a handwritten word-spotting task. The method exploits the fact that the semi-continuous hidden Markov model separates the word model parameters into (i) a codebook of shapes and (ii) a set of word-specific parameters.
Our main contribution is to employ this property to derive writer-specific word models by statistically adapting an initial universal codebook to each document. This process is unsupervised and does not even require the appearance of the keyword(s) in the searched document. Experimental results show an increase in performance when this adaptation technique is applied. To the best of our knowledge, this is the first work dealing with adaptation for word-spotting. The preliminary version of this paper obtained an IBM Best Student Paper Award at the 19th International Conference on Pattern Recognition.

Abstract: A standard way to deal with multi-class categorization problems is by the combination of binary classifiers in a pairwise voting procedure. Recently, this classical approach has been formalized in the Error-Correcting Output Codes (ECOC) framework. In the ECOC framework, the one-versus-one coding demonstrates to achieve higher performance than the rest of coding designs. The binary problems that we train in the one-versus-one strategy are significantly smaller than in the rest of designs, and usually easier to be learnt, taking into account the smaller overlapping between classes. However, a high percentage of the positions coded by zero of the coding matrix, which implies a high sparseness degree, does not codify meta-class membership information. In this paper, we show that using the training data we can redefine without re-training, in a problem-dependent way, the one-versus-one coding matrix so that the new coded information helps the system to increase its generalization capability. Moreover, the new re-coding strategy is generalized to be applied over any binary code. The results over several UCI Machine Learning repository data sets and two real multi-class problems show that performance improvements can be obtained re-coding the classical one-versus-one and Sparse random designs compared to different state-of-the-art ECOC configurations.

Abstract: IF JCR CCIA 1.303 2009 54/103
This article proposes a general extension of the error correcting output codes framework to the online learning scenario. As a result, the final classifier handles the addition of new classes independently of the base classifier used. In particular, this extension supports the use of both online example incremental and batch classifiers as base learners. The extension of the traditional problem independent codings one-versus-all and one-versus-one is introduced. Furthermore, two new codings are proposed, unbalanced online ECOC and a problem dependent online ECOC. This last online coding technique takes advantage of the problem data for minimizing the number of dichotomizers used in the ECOC framework while preserving a high accuracy. These techniques are validated on an online setting of 11 data sets from UCI database and applied to two real machine vision applications: traffic sign recognition and face recognition. As a result, the online ECOC techniques proposed provide a feasible and robust way for handling new classes using any base classifier.

Abstract: The impressive performance of Convolutional Neural Networks (CNNs) when solving different vision problems is shadowed by their black-box nature and our consequent lack of understanding of the representations they build and how these representations are organized. To help understanding these issues, we propose to describe the activity of individual neurons by their Neuron Feature visualization and quantify their inherent selectivity with two specific properties. We explore selectivity indexes for: an image feature (color); and an image label (class membership). Our contribution is a framework to seek or classify neurons by indexing on these selectivity properties. It helps to find color selective neurons, such as a red-mushroom neuron in layer Conv4 or class selective neurons such as dog-face neurons in layer Conv5 in VGG-M, and establishes a methodology to derive other selectivity properties. Indexing on neuron selectivity can statistically draw how features and classes are represented through layers in a moment when the size of trained nets is growing and automatic tools to index neurons can be helpful.

Abstract: This paper presents a new model for the task of scene text visual question answering. In this task questions about a given image can only be answered by reading and understanding scene text. Current state of the art models for this task make use of a dual attention mechanism in which one attention module attends to visual features while the other attends to textual features. A possible issue with this is that it makes difficult for the model to reason jointly about both modalities. To fix this problem we propose a new model that is based on an single attention mechanism that attends to multi-modal features conditioned to the question. The output weights of this attention module over a grid of multi-modal spatial features are interpreted as the probability that a certain spatial location of the image contains the answer text to the given question. Our experiments demonstrate competitive performance in two standard datasets with a model that is faster than previous methods at inference time. Furthermore, we also provide a novel analysis of the ST-VQA dataset based on a human performance study. Supplementary material, code, and data is made available through this link.