https://www2.econ.iastate.edu/tesfatsi/DeepLearningInNeuralNetworksOverview.JSchmidhuber2015.pdf

Deep learning in neural networks

In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First , we highlight convolution with upsampled filters, or ‘atrous convolution’, as a powerful tool in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second , we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third , we improve the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models. The commonly deployed combination of max-pooling and downsampling in DCNNs achieves invariance but has a toll on localization accuracy. We overcome this by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF), which is shown both qualitatively and quantitatively to improve localization performance. Our proposed “DeepLab” system sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 79.7 percent mIOU in the test set, and advances the results on three other datasets: PASCAL-Context, PASCAL-Person-Part, and Cityscapes. All of our code is made publicly available online.

DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain X to a target domain Y in the absence of paired examples. Our goal is to learn a mapping G : X → Y such that the distribution of images from G(X) is indistinguishable from the distribution Y using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping F : Y → X and introduce a cycle consistency loss to push F(G(X)) ≈ X (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach.

Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First, we highlight convolution with upsampled filters, or 'atrous convolution', as a powerful tool in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second, we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third, we improve the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models. The commonly deployed combination of max-pooling and downsampling in DCNNs achieves invariance but has a toll on localization accuracy. We overcome this by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF), which is shown both qualitatively and quantitatively to improve localization performance. Our proposed "DeepLab" system sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 79.7% mIOU in the test set, and advances the results on three other datasets: PASCAL-Context, PASCAL-Person-Part, and Cityscapes. All of our code is made publicly available online.

We present a system that automatically tags videos, i.e. detects high-level semantic concepts like objects or actions in them. To do so, our system does not rely on datasets manually annotated for research purposes. Instead, we propose to use videos from online portals like youtube.com as a novel source of training data, whereas tags provided by users during upload serve as ground truth annotations. This allows our system to learn autonomously by automatically downloading its training set.

The key contribution of this work is a number of large-scale quantitative experiments on real-world online videos, in which we investigate the influence of the individual system components, and how well our tagger generalizes to novel content. Our key results are: (1) Fair tagging results can be obtained by a late fusion of several kinds of visual features. (2) Using more than one keyframe per shot is helpful. (3) To generalize to different video content (e.g., another video portal), the system can be adapted by expanding its training set.

A system that learns to tag videos by watching youtube

Systems (artificial or natural) for visual object recognition are faced with three fundamental problems: the correspondence problem, the problem of representing 3D shape, and the problem of defining a robust similarity measure between images and views of objects.
In this thesis, I address each of these problems: (1) I present a recognition algorithm (RAST) that works efficiently even when no correspondence or grouping information is given; that is, it works in the presence of large amounts of clutter and with very primitive features. (2) I discuss representations of 3D objects as collections of 2D views for the purposes of visual object recognition. Such representations greatly simplify the problems of model acquisition and representing complex shapes. I present theoretical and empirical evidence that this "view-based approximation" is an efficient, robust, and reliable approach to 3D visual object recognition. (3) I present Bayesian and MDL approaches to the similarity problem that may help us build more robust recognition systems.
These results form the basis for a simple, efficient, and robust approach to the geometric aspects of 3D object recognition from 2D image. The results presented in this thesis also strongly suggest that future research directed towards building reliable recognition systems for real world environments must focus on the non-geometric aspects of visual object recognition, such as statistics and scene interpretation. (Copies available exclusively from MIT Libraries, Rm. 14-0551, Cambridge, MA 02139-4307. Ph. 617-253-5668; Fax 617-253-1690.)

/pdf/geometric-aspects-of-visual-object-recognition-cwdfaaaa7v.pdf

Geometric Aspects of Visual Object Recognition

Language models or recognition dictionaries are usually considered an essential step in OCR. However, using a language model complicates training of OCR systems, and it also narrows the range of texts that an OCR system can be used with. Recent results have shown that Long Short-Term Memory (LSTM) based OCR yields low error rates even without language modeling. In this paper, we explore the question to what extent LSTM models can be used for multilingual OCR without the use of language models. To do this, we measure cross-language performance of LSTM models trained on different languages. LSTM models show good promise to be used for language-independent OCR. The recognition errors are very low (around 1%) without using any language model or dictionary correction.

Can we build language-independent OCR using LSTM networks?

Layout analysis is a key component of an OCR system In this paper, we present a layout analysis system for extracting text-lines in reading order from Urdu document images For this purpose, we evaluate an existing system for Roman script text on Urdu documents and describe its methods and the main changes necessary to adapt it to Urdu script The main changes are: 1) the text-line model for Roman script is modified to adapt to Urdu text, 2) reading order of an Urdu document is defined The method is applied to a collection of scanned Urdu documents from various books, magazines, and newspapers The results show high text-line detection accuracy on scanned images of Urdu prose and poetry books and magazines The algorithm also works reasonably well on newspaper images We also identify directions for future work which may further improve the accuracy of the system

https://www.dfki.de/web/forschung/publikationen/renameFileForDownload?filename=publ-INMIC06.pdf&file_id=uploads_37

Layout Analysis of Urdu Document Images

In this paper a modified decision tree algorithm for anomaly detection is presented. During the tree building process, densities for the outlier class are used directly in the split point determination algorithm. No artificial counter-examples have to be sampled from the unknown class, which yields to more precise decision boundaries and a deterministic classification result. Furthermore, the prior of the outlier class can be used to adjust the sensitivity of the anomaly detector. The proposed method combines the advantages of classification trees with the benefit of a more accurate representation of the outliers. For evaluation, we compare our approach with other state-of-the-art anomaly detection algorithms on four standard data sets including the KDD-Cup 99. The results show that the proposed method performs as well as more complex approaches and is even superior on three out of four data sets.

https://www.dfki.de/web/forschung/iwi/publikationen/renameFileForDownload?filename=Anomaly_Detection_Trees.pdf&file_id=uploads_472

Thomas M. Breuel

Papers

A system that learns to tag videos by watching youtube

Geometric Aspects of Visual Object Recognition

Can we build language-independent OCR using LSTM networks?

Layout Analysis of Urdu Document Images

Anomaly detection by combining decision trees and parametric densities