Temporal action localization is an important yet challenging problem. Given a long, untrimmed video consisting of multiple action instances and complex background contents, we need not only to recognize their action categories, but also to localize the start time and end time of each instance. Many state-of-the-art systems use segment-level classifiers to select and rank proposal segments of pre-determined boundaries. However, a desirable model should move beyond segment-level and make dense predictions at a fine granularity in time to determine precise temporal boundaries. To this end, we design a novel Convolutional-De-Convolutional (CDC) network that places CDC filters on top of 3D ConvNets, which have been shown to be effective for abstracting action semantics but reduce the temporal length of the input data. The proposed CDC filter performs the required temporal upsampling and spatial downsampling operations simultaneously to predict actions at the frame-level granularity. It is unique in jointly modeling action semantics in space-time and fine-grained temporal dynamics. We train the CDC network in an end-to-end manner efficiently. Our model not only achieves superior performance in detecting actions in every frame, but also significantly boosts the precision of localizing temporal boundaries. Finally, the CDC network demonstrates a very high efficiency with the ability to process 500 frames per second on a single GPU server. We will update the camera-ready version and publish the source codes online soon.
Visual relations, such as "person ride bike" and "bike next to car", offer a comprehensive scene understanding of an image, and have already shown their great utility in connecting computer vision and natural language. However, due to the challenging combinatorial complexity of modeling subject-predicate-object relation triplets, very little work has been done to localize and predict visual relations. Inspired by the recent advances in relational representation learning of knowledge bases and convolutional object detection networks, we propose a Visual Translation Embedding network (VTransE) for visual relation detection. VTransE places objects in a low-dimensional relation space where a relation can be modeled as a simple vector translation, i.e., subject + predicate $\approx$ object. We propose a novel feature extraction layer that enables object-relation knowledge transfer in a fully-convolutional fashion that supports training and inference in a single forward/backward pass. To the best of our knowledge, VTransE is the first end-to-end relation detection network. We demonstrate the effectiveness of VTransE over other state-of-the-art methods on two large-scale datasets: Visual Relationship and Visual Genome. Note that even though VTransE is a purely visual model, it is still competitive to the Lu's multi-modal model with language priors.
In applications involving matching of image sets, the information from multiple images must be effectively exploited to represent each set. State-of-the-art methods use probabilistic distribution or subspace to model a set and use specific distance measure to compare two sets. These methods are slow to compute and not compact to use in a large scale scenario. Learning-based hashing is often used in large scale image retrieval as they provide a compact representation of each sample and the Hamming distance can be used to efficiently compare two samples. However, most hashing methods encode each image separately and discard knowledge that multiple images in the same set represent the same object or person. We investigate the set hashing problem by combining both set representation and hashing in a single deep neural network. An image set is first passed to a CNN module to extract image features, then these features are aggregated using two types of set feature to capture both set specific and database-wide distribution information. The computed set feature is then fed into a multilayer perceptron to learn a compact binary embedding. Triplet loss is used to train the network by forming set similarity relations using class labels. We extensively evaluate our approach on datasets used for image matching and show highly competitive performance compared to state-of-the-art methods.
EventNet is a large-scale video corpus and event ontology consisting of 500 events associated with event-specific concepts. In order to improve the quality of the current EventNet, we conduct the following steps and introduce EventNet version 1.1: (1) manually verify the correctness of event labels for all videos; (2) remove the YouTube user bias by limiting the maximum number of videos in each event from the same YouTube user as 3; (3) remove the videos which are currently not accessible online; (4) remove the video belonging to multiple event categories. After the above procedure, some events may contain only a small number of videos, and therefore we crawl more videos for those events to ensure every event will contain more than 50 videos. Finally, EventNet version 1.1 contains 67,641 videos, 500 events, and 5,028 event-specific concepts. In addition, we train a Convolutional Neural Network (CNN) model for event classification via fine-tuning AlexNet using EventNet version 1.1. Then we use the trained CNN model to extract FC7 layer feature and train binary classifiers using linear SVM for each event-specific concept. We believe this new version of EventNet will significantly facilitate research in computer vision and multimedia, and will put it online for public downloading in the future.
Residual learning has recently surfaced as an effective means of constructing very deep neural networks for object recognition. However, current incarnations of residual networks do not allow for the modeling and integration of complex relations between closely coupled recognition tasks or across domains. Such problems are often encountered in multimedia applications involving large-scale content recognition. We propose a novel extension of residual learning for deep networks that enables intuitive learning across multiple related tasks using cross-connections called cross-residuals. These cross-residuals connections can be viewed as a form of in-network regularization and enables greater network generalization. We show how cross-residual learning (CRL) can be integrated in multitask networks to jointly train and detect visual concepts across several tasks. We present a single multitask cross-residual network with >40% less parameters that is able to achieve competitive, or even better, detection performance on a visual sentiment concept detection problem normally requiring multiple specialized single-task networks. The resulting multitask cross-residual network also achieves better detection performance by about 10.4% over a standard multitask residual network without cross-residuals with even a small amount of cross-task weighting.
Robotic manipulation of deformable objects is a difficult problem especially because of the complexity of the many different ways an object can deform. Searching such a high dimensional state space makes it difficult to recognize, track, and manipulate deformable objects. In this paper, we introduce a predictive, model-driven approach to address this challenge, using a pre-computed, simulated database of deformable object models. Mesh models of common deformable garments are simulated with the garments picked up in multiple different poses under gravity, and stored in a database for fast and efficient retrieval. To validate this approach, we developed a comprehensive pipeline for manipulating clothing as in a typical laundry task. First, the database is used for category and pose estimation for a garment in an arbitrary position. A fully featured 3D model of the garment is constructed in real-time and volumetric features are then used to obtain the most similar model in the database to predict the object category and pose. Second, the database can significantly benefit the manipulation of deformable objects via non-rigid registration, providing accurate correspondences between the reconstructed object model and the database models. Third, the accurate model simulation can also be used to optimize the trajectories for manipulation of deformable objects, such as the folding of garments. Extensive experimental results are shown for the tasks above using a variety of different clothing.
The impact of culture in visual emotion perception has recently captured the attention of multimedia research. In this study, we pro- vide powerful computational linguistics tools to explore, retrieve and browse a dataset of 16K multilingual affective visual concepts and 7.3M Flickr images. First, we design an effective crowdsourc- ing experiment to collect human judgements of sentiment connected to the visual concepts. We then use word embeddings to repre- sent these concepts in a low dimensional vector space, allowing us to expand the meaning around concepts, and thus enabling insight about commonalities and differences among different languages. We compare a variety of concept representations through a novel evaluation task based on the notion of visual semantic relatedness. Based on these representations, we design clustering schemes to group multilingual visual concepts, and evaluate them with novel metrics based on the crowdsourced sentiment annotations as well as visual semantic relatedness. The proposed clustering framework enables us to analyze the full multilingual dataset in-depth and also show an application on a facial data subset, exploring cultural in- sights of portrait-related affective visual concepts.
This technical report details several improvements to the visual concept detector banks built on images from the Multilingual Visual Sentiment Ontology (MVSO). The detector banks are trained to detect a total of 9,918 sentiment-biased visual concepts from six major languages: English, Spanish, Italian, French, German and Chinese. In the original MVSO release, adjective-noun pair (ANP) detectors were trained for the six languages using an AlexNet-styled architecture by fine-tuning from DeepSentiBank. Here, through a more extensive set of experiments, parameter tuning, and training runs, we detail and release higher accuracy models for detecting ANPs across six languages from the same image pool and setting as in the original release using a more modern architecture, GoogLeNet, providing comparable or better performance with reduced network parameter cost. In addition, since the image pool in MVSO can be corrupted by user noise from social interactions, we partitioned out a sub-corpus of MVSO images based on tag-restricted queries for higher fidelity labels. We show that as a result of these higher fidelity labels, higher performing AlexNet-styled ANP detectors can be trained using the tag-restricted image subset as compared to the models in full corpus. We release all these newly trained models for public research use along with the list of tag-restricted images from the MVSO dataset.
This paper aims for generic instance search from one example where the instance can be an arbitrary object like shoes, not just near-planar and one-sided instances like buildings and logos. First, we evaluate state-of-the-art instance search methods on this problem. We observe that what works for buildings loses its generality on shoes. Second, we propose to use automatically learned category-specific attributes to address the large appearance variations present in generic instance search. Searching among instances from the same category as the query, the category-specific attributes outperform existing approaches by a large margin on shoes and cars and perform on par with the state-of-the-art on buildings. Third, we treat person re-identification as a special case of generic instance search. On the popular VIPeR dataset, we reach state-of-the-art performance with the same method. Fourth, we extend our method to search objects without restriction to the specifically known category. We show that the combination of category-level information and the category-specific attributes is superior to the alternative method combining category-level information with low-level features such as Fisher vector.
We address temporal action localization in untrimmed long videos. This is important because videos in real applications are usually unconstrained and contain multiple action instances plus video content of background scenes or other activities. To address this challenging issue, we exploit the effectiveness of deep networks in temporal action localization via three segment-based 3D ConvNets: (1) a proposal network identifies candidate segments in a long video that may contain actions; (2) a classification network learns one-vs-all action classification model to serve as initialization for the localization network; and (3) a localization network fine-tunes on the learned classification network to localize each action instance. We propose a novel loss function for the localization network to explicitly consider temporal overlap and therefore achieve high temporal localization accuracy. Only the proposal network and the localization network are used during prediction. On two large-scale benchmarks, our approach achieves significantly superior performances compared with other state-of-the-art systems: mAP increases from 1.7% to 7.4% on MEXaction2 and increases from 15.0% to 19.0% on THUMOS 2014, when the overlap threshold for evaluation is set to 0.5.