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.
The accuracy of information retrieval systems is often measured using complex loss functions such as the average precision (AP) or the normalized discounted cumulative gain (NDCG). Given a set of positive and negative samples, the parameters of a retrieval system can be estimated by minimizing these loss functions. However, the non-differentiability and non-decomposability of these loss functions does not allow for simple gradient based optimization algorithms. This issue is generally circumvented by either optimizing a structured hinge-loss upper bound to the loss function or by using asymptotic methods like the direct-loss minimization framework. Yet, the high computational complexity of loss-augmented inference, which is necessary for both the frameworks, prohibits its use in large training data sets. To alleviate this deficiency, we present a novel quicksort flavored algorithm for a large class of non-decomposable loss functions. We provide a complete characterization of the loss functions that are amenable to our algorithm, and show that it includes both AP and NDCG based loss functions. Furthermore, we prove that no comparison based algorithm can improve upon the computational complexity of our approach asymptotically. We demonstrate the effectiveness of our approach in the context of optimizing the structured hinge loss upper bound of AP and NDCG loss for learning models for a variety of vision tasks. We show that our approach provides significantly better results than simpler decomposable loss functions, while requiring a comparable training time.
We present a framework for learning efficient holistic representation for handwritten word images. The proposed method uses a deep convolutional neural network with traditional classification loss. The major strengths of our work lie in: (i) the efficient usage of synthetic data to pre-train a deep network, (ii) an adapted version of ResNet-34 architecture with region of interest pooling (referred as HWNet v2) which learns discriminative features with variable sized word images, and (iii) realistic augmentation of training data with multiple scales and elastic distortion which mimics the natural process of handwriting. We further investigate the process of fine-tuning at various layers to reduce the domain gap between synthetic and real domain and also analyze the in-variances learned at different layers using recent visualization techniques proposed in literature. Our representation leads to state of the art word spotting performance on standard handwritten datasets and historical manuscripts in different languages with minimal representation size. On the challenging IAM dataset, our method is first to report an mAP above 0.90 for word spotting with a representation size of just 32 dimensions. Further more, we also present results on printed document datasets in English and Indic scripts which validates the generic nature of the proposed framework for learning word image representation.
In this paper, we demonstrate a score based indexing approach for tennis videos. Given a broadcast tennis video (BTV), we index all the video segments with their scores to create a navigable and searchable match. Our approach temporally segments the rallies in the video and then recognizes the scores from each of the segments, before refining the scores using the knowledge of the tennis scoring system. We finally build an interface to effortlessly retrieve and view the relevant video segments by also automatically tagging the segmented rallies with human accessible tags such as 'fault' and 'deuce'. The efficiency of our approach is demonstrated on BTV's from two major tennis tournaments.
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.
Building robust recognizers for Arabic has always been challenging. We demonstrate the effectiveness of an end-to-end trainable CNN-RNN hybrid architecture in recognizing Arabic text in videos and natural scenes. We outperform previous state-of-the-art on two publicly available video text datasets - ALIF and ACTIV. For the scene text recognition task, we introduce a new Arabic scene text dataset and establish baseline results. For scripts like Arabic, a major challenge in developing robust recognizers is the lack of large quantity of annotated data. We overcome this by synthesising millions of Arabic text images from a large vocabulary of Arabic words and phrases. Our implementation is built on top of the model introduced here [37] which is proven quite effective for English scene text recognition. The model follows a segmentation-free, sequence to sequence transcription approach. The network transcribes a sequence of convolutional features from the input image to a sequence of target labels. This does away with the need for segmenting input image into constituent characters/glyphs, which is often difficult for Arabic script. Further, the ability of RNNs to model contextual dependencies yields superior recognition results.
The success of deep learning in computer vision has greatly increased the need for annotated image datasets. We propose an EEG (Electroencephalogram)-based image annotation system. While humans can recognize objects in 20-200 milliseconds, the need to manually label images results in a low annotation throughput. Our system employs brain signals captured via a consumer EEG device to achieve an annotation rate of up to 10 images per second. We exploit the P300 event-related potential (ERP) signature to identify target images during a rapid serial visual presentation (RSVP) task. We further perform unsupervised outlier removal to achieve an F1-score of 0.88 on the test set. The proposed system does not depend on category-specific EEG signatures enabling the annotation of any new image category without any model pre-training.
End-to-end training from scratch of current deep architectures for new computer vision problems would require Imagenet-scale datasets, and this is not always possible. In this paper we present a method that is able to take advantage of freely available multi-modal content to train computer vision algorithms without human supervision. We put forward the idea of performing self-supervised learning of visual features by mining a large scale corpus of multi-modal (text and image) documents. We show that discriminative visual features can be learnt efficiently by training a CNN to predict the semantic context in which a particular image is more probable to appear as an illustration. For this we leverage the hidden semantic structures discovered in the text corpus with a well-known topic modeling technique. Our experiments demonstrate state of the art performance in image classification, object detection, and multi-modal retrieval compared to recent self-supervised or natural-supervised approaches.
In this paper, we propose a novel method to register football broadcast video frames on the static top view model of the playing surface. The proposed method is fully automatic in contrast to the current state of the art which requires manual initialization of point correspondences between the image and the static model. Automatic registration using existing approaches has been difficult due to the lack of sufficient point correspondences. We investigate an alternate approach exploiting the edge information from the line markings on the field. We formulate the registration problem as a nearest neighbour search over a synthetically generated dictionary of edge map and homography pairs. The synthetic dictionary generation allows us to exhaustively cover a wide variety of camera angles and positions and reduce this problem to a minimal per-frame edge map matching procedure. We show that the per-frame results can be improved in videos using an optimization framework for temporal camera stabilization. We demonstrate the efficacy of our approach by presenting extensive results on a dataset collected from matches of football World Cup 2014.
Generating synthetic images is an art which emulates the natural process of image generation in a closest possible manner. In this work, we exploit such a framework for data generation in handwritten domain. We render synthetic data using open source fonts and incorporate data augmentation schemes. As part of this work, we release 9M synthetic handwritten word image corpus which could be useful for training deep network architectures and advancing the performance in handwritten word spotting and recognition tasks.