Convolutional Neural Network (CNN) based image segmentation has made great progress in recent years. However, video object segmentation remains a challenging task due to its high computational complexity. Most of the previous methods employ a two-stream CNN framework to handle spatial and motion features separately. In this paper, we propose an end-to-end encoder-decoder style 3D CNN to aggregate spatial and temporal information simultaneously for video object segmentation. To efficiently process video, we propose 3D separable convolution for the pyramid pooling module and decoder, which dramatically reduces the number of operations while maintaining the performance. Moreover, we also extend our framework to video action segmentation by adding an extra classifier to predict the action label for actors in videos. Extensive experiments on several video datasets demonstrate the superior performance of the proposed approach for action and object segmentation compared to the state-of-the-art.
In this work, we propose an end-to-end constrained clustering scheme to tackle the person re-identification (re-id) problem. Deep neural networks (DNN) have recently proven to be effective on person re-identification task. In particular, rather than leveraging solely a probe-gallery similarity, diffusing the similarities among the gallery images in an end-to-end manner has proven to be effective in yielding a robust probe-gallery affinity. However, existing methods do not apply probe image as a constraint, and are prone to noise propagation during the similarity diffusion process. To overcome this, we propose an intriguing scheme which treats person-image retrieval problem as a {\em constrained clustering optimization} problem, called deep constrained dominant sets (DCDS). Given a probe and gallery images, we re-formulate person re-id problem as finding a constrained cluster, where the probe image is taken as a constraint (seed) and each cluster corresponds to a set of images corresponding to the same person. By optimizing the constrained clustering in an end-to-end manner, we naturally leverage the contextual knowledge of a set of images corresponding to the given person-images. We further enhance the performance by integrating an auxiliary net alongside DCDS, which employs a multi-scale Resnet. To validate the effectiveness of our method we present experiments on several benchmark datasets and show that the proposed method can outperform state-of-the-art methods.
The visual entities in cross-view images exhibit drastic domain changes due to the difference in viewpoints each set of images is captured from. Existing state-of-the-art methods address the problem by learning view-invariant descriptors for the images. We propose a novel method for solving this task by exploiting the generative powers of conditional GANs to synthesize an aerial representation of a ground level panorama and use it to minimize the domain gap between the two views. The synthesized image being from the same view as the target image helps the network to preserve important cues in aerial images following our Joint Feature Learning approach. Our Feature Fusion method combines the complementary features from a synthesized aerial image with the corresponding ground features to obtain a robust query representation. In addition, multi-scale feature aggregation preserves image representations at different feature scales useful for solving this complex task. Experimental results show that our proposed approach performs significantly better than the state-of-the-art methods on the challenging CVUSA dataset in terms of top-1 and top-1% retrieval accuracies. Furthermore, to evaluate the generalization of our method on urban landscapes, we collected a new cross-view localization dataset with geo-reference information.
In this paper, we tackle the problem of Crowd Counting, and present a crowd density estimation based approach for obtaining the crowd count. Most of the existing crowd counting approaches rely on local features for estimating the crowd density map. In this work, we investigate the usefulness of combining local with non-local features for crowd counting. We use convolution layers for extracting local features, and a type of self-attention mechanism for extracting non-local features. We combine the local and the non-local features, and use it for estimating crowd density map. We conduct experiments on three publicly available Crowd Counting datasets, and achieve significant improvement over the previous approaches.
Widespread use of wearable cameras and recording devices such as cellphones have opened the door to a lot of interesting research in first-person Point-of-view (POV) videos (egocentric videos). In recent years, we have seen the performance of video-based person Re-Identification (ReID) methods improve considerably. However, with the influx of varying video domains, such as egocentric videos, it has become apparent that there are still many open challenges to be faced. These challenges are a result of factors such as poor video quality due to ego-motion, blurriness, severe changes in lighting conditions and perspective distortions. To facilitate the research towards conquering these challenges, this paper contributes a new, first-of-its-kind dataset called EgoReID. The dataset is captured using 3 mobile cellphones with non-overlapping field-of-view. It contains 900 IDs and around 10,200 tracks with a total of 176,000 detections. Moreover, for each video we also provide 12-sensor meta data. Directly applying current approaches to our dataset results in poor performance. Considering the unique nature of our dataset, we propose a new framework which takes advantage of both visual and sensor meta data to successfully perform Person ReID. In this paper, we propose to adopt human body region parsing to extract local features from different body regions and then employ 3D convolution to better encode temporal information of each sequence of body parts. In addition, we also employ sensor meta data to determine target's next camera and their estimated time of arrival, such that the search is only performed among tracks present in the predicted next camera around the estimated time. This considerably improves our ReID performance as it significantly reduces our search space.
In this paper, we propose an end-to-end capsule network for pixel level localization of actors and actions present in a video. The localization is performed based on a natural language query through which an actor and action are specified. We propose to encode both the video as well as textual input in the form of capsules, which provide more effective representation in comparison with standard convolution based features. We introduce a novel capsule based attention mechanism for fusion of video and text capsules for text selected video segmentation. The attention mechanism is performed via joint EM routing over video and text capsules for text selected actor and action localization. The existing works on actor-action localization are mainly focused on localization in a single frame instead of the full video. Different from existing works, we propose to perform the localization on all frames of the video. To validate the potential of the proposed network for actor and action localization on all the frames of a video, we extend an existing actor-action dataset (A2D) with annotations for all the frames. The experimental evaluation demonstrates the effectiveness of the proposed capsule network for text selective actor and action localization in videos, and it also improves upon the performance of the existing state-of-the art works on single frame-based localization.
The goal of data selection is to capture the most structural information from a set of data. This paper presents a fast and accurate data selection method, in which the selected samples are optimized to span the subspace of all data. We propose a new selection algorithm, referred to as iterative projection and matching (IPM), with linear complexity w.r.t. the number of data, and without any parameter to be tuned. In our algorithm, at each iteration, the maximum information from the structure of the data is captured by one selected sample, and the captured information is neglected in the next iterations by projection on the null-space of previously selected samples. The computational efficiency and the selection accuracy of our proposed algorithm outperform those of the conventional methods. Furthermore, the superiority of the proposed algorithm is shown on active learning for video action recognition dataset on UCF-101; learning using representatives on ImageNet; training a generative adversarial network (GAN) to generate multi-view images from a single-view input on CMU Multi-PIE dataset; and video summarization on UTE Egocentric dataset.
The recent success in deep learning has lead to various effective representation learning methods for videos. However, the current approaches for video representation require large amount of human labeled datasets for effective learning. We present an unsupervised representation learning framework to encode scene dynamics in videos captured from multiple viewpoints. The proposed framework has two main components: Representation Learning Network (RL-NET), which learns a representation with the help of Blending Network (BL-NET), and Video Rendering Network (VR-NET), which is used for video synthesis. The framework takes as input video clips from different viewpoints and time, learns an internal representation and uses this representation to render a video clip from an arbitrary given viewpoint and time. The ability of the proposed network to render video frames from arbitrary viewpoints and time enable it to learn a meaningful and robust representation of the scene dynamics. We demonstrate the effectiveness of the proposed method in rendering view-aware as well as time-aware video clips on two different real-world datasets including UCF-101 and NTU-RGB+D. To further validate the effectiveness of the learned representation, we use it for the task of view-invariant activity classification where we observe a significant improvement (~26%) in the performance on NTU-RGB+D dataset compared to the existing state-of-the art methods.
Several recent studies have demonstrated the promise of deep visuomotor policies for robot manipulator control. Despite impressive progress, these systems are known to be vulnerable to physical disturbances, such as accidental or adversarial bumps that make them drop the manipulated object. They also tend to be distracted by visual disturbances such as objects moving in the robot's field of view, even if the disturbance does not physically prevent the execution of the task. In this paper, we propose an approach for augmenting a deep visuomotor policy trained through demonstrations with Task Focused visual Attention (TFA). The manipulation task is specified with a natural language text such as `move the red bowl to the left'. This allows the visual attention component to concentrate on the current object that the robot needs to manipulate. We show that even in benign environments, the TFA allows the policy to consistently outperform a variant with no attention mechanism. More importantly, the new policy is significantly more robust: it regularly recovers from severe physical disturbances (such as bumps causing it to drop the object) from which the baseline policy, i.e. with no visual attention, almost never recovers. In addition, we show that the proposed policy performs correctly in the presence of a wide class of visual disturbances, exhibiting a behavior reminiscent of human selective visual attention experiments. Our proposed approach consists of a VAE-GAN network which encodes the visual input and feeds it to a Motor network that moves the robot joints. Also, our approach benefits from a teacher network for the TFA that leverages textual input command to robustify the visual encoder against various types of disturbances.
Few-shot or one-shot learning of classifiers for images or videos is an important next frontier in computer vision. The extreme paucity of training data means that the learning must start with a significant inductive bias towards the type of task to be learned. One way to acquire this is by meta-learning on tasks similar to the target task. However, if the meta-learning phase requires labeled data for a large number of tasks closely related to the target task, it not only increases the difficulty and cost, but also conceptually limits the approach to variations of well-understood domains. In this paper, we propose UMTRA, an algorithm that performs meta-learning on an unlabeled dataset in an unsupervised fashion, without putting any constraint on the classifier network architecture. The only requirements towards the dataset are: sufficient size, diversity and number of classes, and relevance of the domain to the one in the target task. Exploiting this information, UMTRA generates synthetic training tasks for the meta-learning phase. We evaluate UMTRA on few-shot and one-shot learning on both image and video domains. To the best of our knowledge, we are the first to evaluate meta-learning approaches on UCF-101. On the Omniglot and Mini-Imagenet few-shot learning benchmarks, UMTRA outperforms every tested approach based on unsupervised learning of representations, while alternating for the best performance with the recent CACTUs algorithm. Compared to supervised model-agnostic meta-learning approaches, UMTRA trades off some classification accuracy for a vast decrease in the number of labeled data needed. For instance, on the five-way one-shot classification on the Omniglot, we retain 85% of the accuracy of MAML, a recently proposed supervised meta-learning algorithm, while reducing the number of required labels from 24005 to 5.