Abstract

The success of deep learning based models have centered around recent architectures and the availability of large scale annotated data. In this work, we explore these two factors systematically for improving handwritten recognition for scanned off-line document images. We propose a modified CNN-RNN hybrid architecture with a major focus on effective training using: (i) efficient initialization of network using synthetic data for pretraining, (ii) image normalization for slant correction and (iii) domain specific data transformation and distortion for learning important invariances. We perform a detailed ablation study to analyze the contribution of individual modules and present state of art results for the task of unconstrained line and word recognition on popular datasets such as IAM, RIMES and GW.

Abstract

Abstract—Handwriting recognition (HWR) in Indic scripts is a challenging problem due to the inherent subtleties in the scripts, cursive nature of the handwriting and similar shape of the characters. Lack of publicly available handwriting datasets in Indic scripts has affected the development of handwritten word recognizers, and made direct comparisons across different methods an impossible task in the field. In this paper, we propose a framework for annotating large scale of handwritten word images with ease and speed. We also release a new handwritten word dataset for Telugu, which is collected and annotated using the proposed framework. We also benchmark major Indic scripts such as Devanagari, Bangla and Telugu for the tasks of word spotting and handwriting recognition using state of the art deep neural architectures. Finally, we evaluate the proposed pipeline on RoyDB, a public dataset, and achieve significant reduction in error rates.

Abstract

Lecture videos are rich with textual information and to be able to understand the text is quite useful for larger video understanding/analysis applications. Though text recognition from images have been an active research area in computer vision, text in lecture videos has mostly been overlooked. In this work, we investigate the efficacy of state-of-the art handwritten and scene text recognition methods on text in lecture videos. To this end, a new dataset - LectureVideoDB compiled from frames from multiple lecture videos is introduced. Our experiments show that the existing methods do not fare well on the new dataset. The results necessitate the need to improvise the existing methods for robust performance on lecture videos.

Abstract

We are motivated by the fact that multiple representations of the environment are required to stand for the changes in appearance with time and for changes that appear in a cyclic manner. These changes are, for example, from day to night time, and from day to day across seasons. In such situations, the robot visits the same routes multiple times and collects different appearances of it. Multiple visual experiences usually find robotic vision applications like visual localization, mapping, place recognition, and autonomous navigation. The novelty in this paper is an algorithm that connects multiple visual experiences via aligning multiple image sequences. This problem is solved by finding the maximum flow in a directed graph flow-network, whose vertices represent the matches between frames in the test and reference sequences. Edges of the graph represent the cost of these matches. The problem of finding the best match is reduced to finding the minimum-cut surface, which is solved as a maximum flow network problem. Application to visual localization is considered in this paper to show the effectiveness of the proposed multiple image sequence alignment method, without loosing its generality. Experimental evaluations show that the precision of sequence matching is improved by considering multiple visual sequences for the same route, and that the method performs favorably against state-of-the-art single representation methods like SeqSLAM and ABLE-M.

Abstract

We consider the task of learning to estimate human pose in still images. In order to avoid the high cost of full supervision, we propose to use a diverse data set, which consists of two types of annotations: (i) a small number of images are labeled using the expensive ground-truth pose; and (ii) other images are labeled using the inexpensive action label. As action information helps narrow down the pose of a human, we argue that this approach can help reduce the cost of training without significantly affecting the accuracy. To demonstrate this we design a probabilistic framework that employs two distributions: (i) a conditional distribution to model the uncertainty over the human pose given the image and the action; and (ii) a prediction distribution, which provides the pose of an image without using any action information. We jointly estimate the parameters of the two aforementioned distributions by minimizing their dissimilarity coefficient, as measured by a task-specific loss function. During both training and testing, we only require an efficient sampling strategy for both the aforementioned distributions. This allows us to use deep probabilistic networks that are capable of providing accurate pose estimates for previously unseen images. Using the MPII data set, we show that our approach outperforms baseline methods that either do not use the diverse annotations or rely on pointwise estimates of the pose.

Abstract

The immense success of deep learning based methods in computer vision heavily relies on large scale training datasets. These richly annotated datasets help the network learn discriminative visual features. Collecting and annotating such datasets requires a tremendous amount of human effort and annotations are limited to popular set of classes. As an alternative, learning visual features by designing auxiliary tasks which make use of freely available self-supervision has become increasingly popular in the computer vision community. In this paper, we put forward an idea to take advantage of multi-modal context to provide self-supervision for the training of computer vision algorithms. We show that adequate visual features can be learned efficiently by training a CNN to predict the semantic textual context in which a particular image is more probable to appear as an illustration. More specifically we use popular text embedding techniques to provide the self-supervision for the training of deep CNN.

Abstract

While dealing with multi-modal data such as pairs of images and text, though individual samples may demonstrate inherent heterogeneity in their content, they are usually coupled with each other based on some higher-level concepts such as their categories. This shared information can be useful in measuring semantics of samples across modalities in a relative manner. In this paper, we investigate the problem of analyzing the degree of specificity in the semantic content of a sample in one modality with respect to semantically similar samples in another modality. Samples that have high similarity with semantically similar samples from another modality are considered to be specific, while others are considered to be relatively ambiguous. To model this property, we propose a novel notion of “cross-specificity”. We present two mechanisms to measure cross-specificity: one based on human judgment and other based on an automated approach. We analyze different aspects of cross-specificity, and demonstrate its utility in cross-modal retrieval task. Experiments show that though conceptually simple, it can benefit several existing cross-modal retrieval techniques, and provides significant boost in their performance.

Abstract

We present an approach for unsupervised training of CNNs in order to learn discriminative face representations. We mine supervised training data by noting that multiple faces in the same video frame must belong to different persons and the same face tracked across multiple frames must belong to the same person. We obtain millions of face pairs from hundreds of videos without using any manual supervision. Although faces extracted from videos have a lower spatial resolution than those which are available as part of standard supervised face datasets such as LFW and CASIA-WebFace, the former represent a much more realistic setting, e.g. in surveillance scenarios where most of the faces detected are very small. We train our CNNs with the relatively low resolution faces extracted from video frames collected, and achieve a higher verification accuracy on the benchmark LFW dataset cf. hand-crafted features such as LBPs, and even surpasses the performance of state-of-the-art deep networks such as VGG-Face, when they are made to work with low resolution input images.

Abstract

Deep convolutional features for word images and textual embedding schemes have shown great success in word spotting. In this work, we follow these motivations to propose an End2End embedding framework which jointly learns both the text and image embeddings using state of the art deep convolutional architectures. The three major contributions of this work are: (i) an End2End embedding scheme to learn a common representation for word images and its labels, (ii) building a state of art word image descriptor and demonstrating its utility as off-the-shelf features for word spotting, and (iii) use of synthetic data as a complementary modality to further enhance word spotting and recognition. On the challenging IAM handwritten dataset, we report a mAP of 0.9509 for query-by-string based retrieval task. Under lexicon based word recognition, our proposed method report a 2.66 and 5.10 CER and WER respectively.

Abstract

Sports video data is recorded for nearly every major tournament but remains archived and inaccessible to large scale data mining and analytics. It can only be viewed sequentially or manually tagged with higher-level labels which is time consuming and prone to errors. In this work, we propose an end-to-end framework for automatic attributes tagging and analysis of sport videos. We use commonly available broadcast videos of matches and, unlike previous approaches, does not rely on special camera setups or additional sensors. Our focus is on Badminton as the sport of interest. We propose a method to analyze a large corpus of badminton broadcast videos by segmenting the points played, tracking and recognizing the players in each point and annotating their respective badminton strokes. We evaluate the performance on 10 Olympic matches with 20 players and achieved 95.44% point segmentation accuracy, 97.38% player detection score (mAP@0.5), 97.98% player identification accuracy, and stroke segmentation edit scores of 80.48%. We further show that the automatically annotated videos alone could enable the gameplay analysis and inference by computing understandable metrics such as player’s reaction time, speed, and footwork around the court, etc.