Abstract:Face detection is frequently attempted by using heavy pre-trained backbone networks like ResNet-50/101/152 and VGG16/19. Few recent works have also proposed lightweight detectors with customized backbones, novel loss functions and efficient training strategies. The novelty of this work lies in the design of a lightweight detector while training with only the commonly used loss functions and learning strategies. The proposed face detector grossly follows the established RetinaFace architecture. The first contribution of this work is the design of a customized lightweight backbone network (BLite) having 0.167M parameters with 0.52 GFLOPs. The second contribution is the use of two independent multi-task losses. The proposed lightweight face detector (FDLite) has 0.26M parameters with 0.94 GFLOPs. The network is trained on the WIDER FACE dataset. FDLite is observed to achieve 92.3\%, 89.8\%, and 82.2\% Average Precision (AP) on the easy, medium, and hard subsets of the WIDER FACE validation dataset, respectively.
Abstract:Underwater image quality is affected by fluorescence, low illumination, absorption, and scattering. Recent works in underwater image enhancement have proposed different deep network architectures to handle these problems. Most of these works have proposed a single network to handle all the challenges. We believe that deep networks trained for specific conditions deliver better performance than a single network learned from all degradation cases. Accordingly, the first contribution of this work lies in the proposal of an iterative framework where a single dominant degradation condition is identified and resolved. This proposal considers the following eight degradation conditions -- low illumination, low contrast, haziness, blurred image, presence of noise and color imbalance in three different channels. A deep network is designed to identify the dominant degradation condition. Accordingly, an appropriate deep network is selected for degradation condition-specific enhancement. The second contribution of this work is the construction of degradation condition specific datasets from good quality images of two standard datasets (UIEB and EUVP). This dataset is used to learn the condition specific enhancement networks. The proposed approach is found to outperform nine baseline methods on UIEB and EUVP datasets.
Abstract:The use of complex attention modules has improved the performance of the Visual Question Answering (VQA) task. This work aims to learn an improved multi-modal representation through dense interaction of visual and textual modalities. The proposed model has an attention block containing both self-attention and co-attention on image and text. The self-attention modules provide the contextual information of objects (for an image) and words (for a question) that are crucial for inferring an answer. On the other hand, co-attention aids the interaction of image and text. Further, fine-grained information is obtained from two modalities by using a Cascade of Self- and Co-Attention blocks (CSCA). This proposal is benchmarked on the widely used VQA2.0 and TDIUC datasets. The efficacy of key components of the model and cascading of attention modules are demonstrated by experiments involving ablation analysis.
Abstract:Whenever we speak, our voice is accompanied by facial movements and expressions. Several recent works have shown the synthesis of highly photo-realistic videos of talking faces, but they either require a source video to drive the target face or only generate videos with a fixed head pose. This lack of facial movement is because most of these works focus on the lip movement in sync with the audio while assuming the remaining facial keypoints' fixed nature. To address this, a unique audio-keypoint dataset of over 150,000 videos at 224p and 25fps is introduced that relates the facial keypoint movement for the given audio. This dataset is then further used to train the model, Audio2Keypoint, a novel approach for synthesizing facial keypoint movement to go with the audio. Given a single image of the target person and an audio sequence (in any language), Audio2Keypoint generates a plausible keypoint movement sequence in sync with the input audio, conditioned on the input image to preserve the target person's facial characteristics. To the best of our knowledge, this is the first work that proposes an audio-keypoint dataset and learns a model to output the plausible keypoint sequence to go with audio of any arbitrary length. Audio2Keypoint generalizes across unseen people with a different facial structure allowing us to generate the sequence with the voice from any source or even synthetic voices. Instead of learning a direct mapping from audio to video domain, this work aims to learn the audio-keypoint mapping that allows for in-plane and out-of-plane head rotations, while preserving the person's identity using a Pose Invariant (PIV) Encoder.
Abstract:Even though there has been tremendous progress in the field of Visual Question Answering, models today still tend to be inconsistent and brittle. To this end, we propose a model-independent cyclic framework which increases consistency and robustness of any VQA architecture. We train our models to answer the original question, generate an implication based on the answer and then also learn to answer the generated implication correctly. As a part of the cyclic framework, we propose a novel implication generator which can generate implied questions from any question-answer pair. As a baseline for future works on consistency, we provide a new human annotated VQA-Implications dataset. The dataset consists of ~30k questions containing implications of 3 types - Logical Equivalence, Necessary Condition and Mutual Exclusion - made from the VQA v2.0 validation dataset. We show that our framework improves consistency of VQA models by ~15% on the rule-based dataset, ~7% on VQA-Implications dataset and robustness by ~2%, without degrading their performance. In addition, we also quantitatively show improvement in attention maps which highlights better multi-modal understanding of vision and language.
Abstract:This paper proposes CQ-VQA, a novel 2-level hierarchical but end-to-end model to solve the task of visual question answering (VQA). The first level of CQ-VQA, referred to as question categorizer (QC), classifies questions to reduce the potential answer search space. The QC uses attended and fused features of the input question and image. The second level, referred to as answer predictor (AP), comprises of a set of distinct classifiers corresponding to each question category. Depending on the question category predicted by QC, only one of the classifiers of AP remains active. The loss functions of QC and AP are aggregated together to make it an end-to-end model. The proposed model (CQ-VQA) is evaluated on the TDIUC dataset and is benchmarked against state-of-the-art approaches. Results indicate competitive or better performance of CQ-VQA.
Abstract:Deep Reinforcement Learning has enabled the learning of policies for complex tasks in partially observable environments, without explicitly learning the underlying model of the tasks. While such model-free methods achieve considerable performance, they often ignore the structure of task. We present a natural representation of to Reinforcement Learning (RL) problems using Recurrent Convolutional Neural Networks (RCNNs), to better exploit this inherent structure. We define 3 such RCNNs, whose forward passes execute an efficient Value Iteration, propagate beliefs of state in partially observable environments, and choose optimal actions respectively. Backpropagating gradients through these RCNNs allows the system to explicitly learn the Transition Model and Reward Function associated with the underlying MDP, serving as an elegant alternative to classical model-based RL. We evaluate the proposed algorithms in simulation, considering a robot planning problem. We demonstrate the capability of our framework to reduce the cost of replanning, learn accurate MDP models, and finally re-plan with learnt models to achieve near-optimal policies.
Abstract:The text data present in overlaid bands convey brief descriptions of news events in broadcast videos. The process of text extraction becomes challenging as overlay text is presented in widely varying formats and often with animation effects. We note that existing edge density based methods are well suited for our application on account of their simplicity and speed of operation. However, these methods are sensitive to thresholds and have high false positive rates. In this paper, we present a contrast enhancement based preprocessing stage for overlay text detection and a parameter free edge density based scheme for efficient text band detection. The second contribution of this paper is a novel approach for multiple text region tracking with a formal identification of all possible detection failure cases. The tracking stage enables us to establish the temporal presence of text bands and their linking over time. The third contribution is the adoption of Tesseract OCR for the specific task of overlay text recognition using web news articles. The proposed approach is tested and found superior on news videos acquired from three Indian English television news channels along with benchmark datasets.
Abstract:Commercial detection in news broadcast videos involves judicious selection of meaningful audio-visual feature combinations and efficient classifiers. And, this problem becomes much simpler if these combinations can be learned from the data. To this end, we propose an Multiple Kernel Learning based method for boosting successful kernel functions while ignoring the irrelevant ones. We adopt a intermediate fusion approach where, a SVM is trained with a weighted linear combination of different kernel functions instead of single kernel function. Each kernel function is characterized by a feature set and kernel type. We identify the feature sub-space locations of the prediction success of a particular classifier trained only with particular kernel function. We propose to estimate a weighing function using support vector regression (with RBF kernel) for each kernel function which has high values (near 1.0) where the classifier learned on kernel function succeeded and lower values (nearly 0.0) otherwise. Second contribution of this work is TV News Commercials Dataset of 150 Hours of News videos. Classifier trained with our proposed scheme has outperformed the baseline methods on 6 of 8 benchmark dataset and our own TV commercials dataset.
Abstract:This paper looks into the problem of pedestrian tracking using a monocular, potentially moving, uncalibrated camera. The pedestrians are located in each frame using a standard human detector, which are then tracked in subsequent frames. This is a challenging problem as one has to deal with complex situations like changing background, partial or full occlusion and camera motion. In order to carry out successful tracking, it is necessary to resolve associations between the detected windows in the current frame with those obtained from the previous frame. Compared to methods that use temporal windows incorporating past as well as future information, we attempt to make decision on a frame-by-frame basis. An occlusion reasoning scheme is proposed to resolve the association problem between a pair of consecutive frames by using an affinity matrix that defines the closeness between a pair of windows and then, uses a binary integer programming to obtain unique association between them. A second stage of verification based on SURF matching is used to deal with those cases where the above optimization scheme might yield wrong associations. The efficacy of the approach is demonstrated through experiments on several standard pedestrian datasets.