Abstract

Diabetic retinopathy (DR) is a disease which is widely diagnosed using (colour fundus) images. Efficiency and accuracy are critical in diagnosing DR as lack of timely intervention can lead to irreversible visual impairment. In this paper, we examine strategies for scrutinizing images which affect diagnostic performance of medical practitioners via an eye-tracking study. A total of 56 subjects with 0 to 18 years of experience participated in the study. Every subject was asked to detect DR from 40 images. The findings indicate that practitioners use mainly two types of strategies characterized by either higher dwell duration or longer track length. The main findings of the study are that higher dwell-based strategy led to higher average accuracy (> 85%) in diagnosis, irrespective of the expertise of practitioner; whereas, the average obtained accuracy with a long-track length-based strategy was dependent on the expertise of the practitioner. In the second part of the paper, we use the experimental findings to recommend a scanning strategy for fast and accurate diagnosis of DR that can be potentially used by image readers. This is derived by combining the eyetracking gaze maps of medical experts in a novel manner based on a set of rules. This strategy requires scrutiny of images in a manner which is consistent with spatial preferences found in human perception in general and in the domain of fundus images in particular. The Levenshtein distance-based assessment of gaze patterns also establish the effectiveness of the derived scanning pattern and is thus recommended for image readers.

Abstract

Annotations are critical for machine learning and developing computer aided diagnosis (CAD) algorithms. Good performance of CAD is critical to their adoption, which generally rely on training with a wide variety of annotated data. However, a vast amount of medical data is either unlabeled or annotated only at the image-level. This poses a problem for exploring data driven approaches like deep learning for CAD. In this paper, we propose a novel crowdsourcing and synthetic image generation for training deep neural net-based lesion detection. The noisy nature of crowdsourced annotations is overcome by assigning a reliability factor for crowd subjects based on their performance and requiring region of interest markings from the crowd. A generative adversarial network-based solution is proposed to generate synthetic images with lesions to control the overall severity level of the disease. We demonstrate the reliability of the crowdsourced annotations and synthetic images by presenting a solution for training the deep neural network (DNN) with data drawn from a heterogeneous mixture of annotations. Experimental results obtained for hard exudate detection from retinal images show that training with refined crowdsourced data/synthetic images is effective as detection performance in terms of sensitivity improves by 25%/27% over training with just expert-markings.

Abstract

Optical coherence tomography (OCT) images are corrupted by speckle noise due to the underlying coherence-based strategy. Speckle suppression/removal in OCT images plays a significant role in both manual and automatic detection of diseases, especially in early clinical diagnosis. In this paper, we propose a new method for denoising OCT images based on Convolutional Neural Network by learning common features from unpaired noisy and clean OCT images in an unsupervised, end-to-end manner. The proposed method consists of a combination of two autoencoders with shared encoder layers, which we call as Shared Encoder (SE) architecture. The SE is trained to reconstruct noisy and clean OCT images with respective autoencoders. The denoised OCT image is obtained using a cross-model prediction. The proposed method can be used for denoising OCT images with or without pathology from any scanner. The SE architecture was assessed using public datasets and found to perform better than baseline methods exhibiting a good balance of retaining anatomical integrity and speckle reduction.

Abstract

Domain Specific Search Engines are expected to provide relevant search results. Availability of enormous number of URLs across subdomains improves relevance of domain specific search engines. The current methods for seed URLs can be systematic ensuring representation of subdomains. We propose a fine grained approach for automatic extraction of seed URLs at subdomain level using Wikipedia and Twitter as repositories. A SeedRel metric and a Diversity Index for seed URL relevance are proposed to measure subdomain coverage. We implemented our approach for 'Security - Information and Cyber' domain and identified 34,007 Seed URLs and 400,726 URLs across subdomains. The measured Diversity index value of 2.10 conforms that all subdomains are represented, hence, a relevant 'Security Search Engine' can be built. Our approach also extracted more URLs (seed and child) as compared to existing approaches for URL extraction.

Abstract

Translation Quality Estimation (QE) aims to estimate the quality of an automated machine translation (MT) output without any human intervention or reference translation. With the increasing use of MT systems in various cross-lingual applications, the need and applicability of QE systems is increasing. We study existing approaches and propose multiple neural network approaches for sentence-level QE, with a focus on MT outputs in Indian languages. For this, we also introduce five new datasets for four language pairs: two for English–Gujarati, and one each for English–Hindi, English–Telugu and English–Bengali, which includes one manually post-edited dataset for English– Gujarati. These Indian languages are spoken by around 689M speakers world-wide. We compare results obtained using our proposed models with multiple state-ofthe-art systems including the winning system in the WMT17 shared task on QE and show that our proposed neural model which combines the discriminative power of carefully chosen features with Siamese Convolutional Neural Networks (CNNs) works best for all Indian language datasets.

Abstract

In order to expand their reach and increase website ad revenue, media outlets have started using clickbait techniques to lure readers to click on articles on their digital platform. Having successfully enticed the user to open the article, the article fails to satiate his curiosity serving only to boost click-through rates. Initial methods for this task were dependent on feature engineering, which varies with each dataset. Industry systems have relied on an exhaustive set of rules to get the job done. Neural networks have barely been explored to perform this task. We propose a novel approach considering different textual embeddings of a news headline and the related article. We generate sub-word level embeddings of the title using Convolutional Neural Networks and use them to train a bidirectional LSTM architecture. An attention layer allows for calculation of significance of each term towards the nature of the post. We also generate Doc2Vec embeddings of the title and article text and model how they interact, following which it is concatenated with the output of the previous component. Finally, this representation is passed through a neural network to obtain a score for the headline. We test our model over 2538 posts (having trained it on 17000 records) and achieve an accuracy of 83.49% outscoring previous state-of-the-art approaches.

Abstract

Localizing natural language phrases in images is a challenging problem that requires joint understanding of both the textual and visual modalities. In the unsupervised setting, lack of supervisory signals exacerbate this difficulty. In this paper, we propose a novel framework for unsupervised visual grounding which uses concept learning as a proxy task to obtain self-supervision. The intuition behind this idea is to encourage the model to localize to regions which can explain some semantic property in the data, in our case, the property being the presence of a concept in a set of images. We present thorough quantitative and qualitative experiments to demonstrate the efficacy of our approach and show a 5.6% improvement over the current state of the art on Visual Genome dataset, a 5.8% improvement on the ReferItGame dataset and comparable to state-of-art performance on the Flickr30k dataset

Abstract

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.

Abstract

Abstract— We present here, a novel network architecture called MergeNet for discovering small obstacles for on-road scenes in the context of autonomous driving. The basis of the architecture rests on the central consideration of training with less amount of data since the physical setup and the annotation process for small obstacles is hard to scale. For making effective use of the limited data, we propose a multi-stage training procedure involving weight-sharing, separate learning of low and high level features from the RGBD input and a refining stage which learns to fuse the obtained complementary features. The model is trained and evaluated on the Lost and Found dataset and is able to achieve state-of-art results with just 135 images in comparison to the 1000 images used by the previous benchmark. Additionally, we also compare our results with recent methods trained on 6000 images and show that our method achieves comparable performance with only 1000 training samples.

Abstract

With news stories coming from a variety of sources, it is crucial for news aggregators to present interesting articles to the user to maximize their engagement. This creates the need for a news recommendation system which understands the content of the articles as well as accounts for the users’ preferences. Methods such as Collaborative Filtering, which are well known for general recommendations, are not suitable for news because of the short life span of articles and because of the large number of articles published each day. Apart from this, such methods do not harness the information present in the sequence in which the articles are read by the user and hence are unable to account for the specific and generic interests of the user which may keep changing with time. In order to address these issues for news recommendation, we propose the Recurrent Attentive Recommendation Engine (RARE). RARE consists of two components and utilizes the distributed representations of news articles. The first component is used to model the user’s sequential behaviour of news reading in order to understand her general interests, i.e., to get a summary of her interests. The second component utilizes an article level attention mechanism to understand her specific interests. We feed the information obtained from both the components to a Siamese Network in order to make predictions which pertain to the user’s generic as well as specific interests. We carry out extensive experiments over three real-world datasets and show that RARE outperforms the state-of-the-art. Furthermore, we also demonstrate the effectiveness of our method in handling the cold start cases.