Temporal Activity Detection aims to predict activity classes per frame, in contrast to video-level predictions in Activity Classification (i.e., Activity Recognition). Due to the expensive frame-level annotations required for detection, the scale of detection datasets is limited. Thus, commonly, previous work on temporal activity detection resorts to fine-tuning a classification model pretrained on large-scale classification datasets (e.g., Kinetics-400). However, such pretrained models are not ideal for downstream detection, due to the disparity between the pretraining and the downstream fine-tuning tasks. In this work, we propose a novel weakly-guided self-supervised pretraining method for detection. We leverage weak labels (classification) to introduce a self-supervised pretext task (detection) by generating frame-level pseudo labels, multi-action frames, and action segments. Simply put, we design a detection task similar to downstream, on large-scale classification data, without extra annotations. We show that the models pretrained with the proposed weakly-guided self-supervised detection task outperform prior work on multiple challenging activity detection benchmarks, including Charades and MultiTHUMOS. Our extensive ablations further provide insights on when and how to use the proposed models for activity detection. Code is available at github.com/kkahatapitiya/SSDet.

Video action detection aims to localize persons in both space and time from video sequences and recognize their actions. Most existing methods are composed of many specialized components, e.g., pretrained person/object detectors, region proposal networks (RPN), memory banks, and so on. This paper proposes a conceptually simple paradigm for video action detection using Transformers, which effectively removes the need for specialized components and achieves superior performance. Our proposed Transformer-based Video Action Detector (TxVAD) utilizes two Transformers to capture scene context information and long-range spatio-temporal context information, for person localization and action classification, respectively. Through extensive experiments on four public datasets, AVA, AVA-Kinetics, JHMDB-21, and UCF101-24, we show that our conceptually simple paradigm has achieved state-of-the-art performance for video action detection task, without using pre-trained person/object detectors, RPN, or memory bank.

Multi-person pose estimation and tracking serve as crucial steps for video understanding. Most state-of-the-art approaches rely on first estimating poses in each frame and only then implementing data association and refinement. Despite the promising results achieved, such a strategy is inevitably prone to missed detections especially in heavily-cluttered scenes, since this tracking-by-detection paradigm is, by nature, largely dependent on visual evidences that are absent in the case of occlusion. In this paper, we propose a novel online approach to learning the pose dynamics, which are independent of pose detections in current fame, and hence may serve as a robust estimation even in challenging scenarios including occlusion. Specifically, we derive this prediction of dynamics through a graph neural network (GNN) that explicitly accounts for both spatial-temporal and visual information. It takes as input the historical pose tracklets and directly predicts the corresponding poses in the following frame for each tracklet. The predicted poses will then be aggregated with the detected poses, if any, at the same frame so as to produce the final pose, potentially recovering the occluded joints missed by the estimator. Experiments on PoseTrack 2017 and Pose-Track 2018 datasets demonstrate that the proposed method achieves results superior to the state of the art on both human pose estimation and tracking tasks.

Visual captioning, the task of describing an image or a video using one or few sentences, is a challenging task owing to the complexity of understanding the copious visual information and describing it using natural language. Motivated by the success of applying neural networks for machine translation, previous work applies sequence to sequence learning to translate videos into sentences. In this work, different from previous work that encodes visual information using a single flow, we introduce a novel Sibling Convolutional Encoder (SibNet) for visual captioning, which employs a dual-branch architecture to collaboratively encode videos. The first content branch encodes visual content information of the video with an autoencoder, capturing the visual appearance information of the video as other networks often do. While the second semantic branch encodes semantic information of the video via visual-semantic joint embedding, which brings complementary representation by considering the semantics when extracting features from videos. Then both branches are effectively combined with soft-attention mechanism and finally fed into a RNN decoder to generate captions. With our SibNet explicitly capturing both content and semantic information, the proposed model can better represent the rich information in videos. To validate the advantages of the proposed model, we conduct experiments on two benchmarks for video captioning, YouTube2Text and MSR-VTT. Our results demonstrate that the proposed SibNet consistently outperforms existing methods across different evaluation metrics.

In this article, we propose a novel object detection algorithm named "Deep Regionlets" by integrating deep neural networks and a conventional detection schema for accurate generic object detection. Motivated by the effectiveness of regionlets for modeling object deformations and multiple aspect ratios, we incorporate regionlets into an end-to-end trainable deep learning framework. The deep regionlets framework consists of a region selection network and a deep regionlet learning module. Specifically, given a detection bounding box proposal, the region selection network provides guidance on where to select sub-regions from which features can be learned from. An object proposal typically contains three - 16 sub-regions. The regionlet learning module focuses on local feature selection and transformations to alleviate the effects of appearance variations. To this end, we first realize non-rectangular region selection within the detection framework to accommodate variations in object appearance. Moreover, we design a "gating network" within the regionlet leaning module to enable instance dependent soft feature selection and pooling. The Deep Regionlets framework is trained end-to-end without additional efforts. We present ablation studies and extensive experiments on the PASCAL VOC dataset and the Microsoft COCO dataset. The proposed method yields competitive performance over state-of-the-art algorithms, such as RetinaNet and Mask R-CNN, even without additional segmentation labels.

Object detection is an essential step towards holistic scene understanding. Most existing object detection algorithms attend to certain object areas once and then predict the object locations. However, scientists have revealed that human do not look at the scene in fixed steadiness. Instead, human eyes move around, locating informative parts to understand the object location. This active perceiving movement process is called saccade. In this paper, inspired by such mechanism, we propose a fast and accurate object detector called SaccadeNet. It contains four main modules, the Center Attentive Module, the Corner Attentive Module, the Attention Transitive Module, and the Aggregation Attentive Module, which allows it to attend to different informative object keypoints actively, and predict object locations from coarse to fine. The Corner Attentive Module is used only during training to extract more informative corner features which brings free-lunch performance boost. On the MS COCO dataset, we achieve the performance of 40.4% mAP at 28 FPS and 30.5% mAP at 118 FPS. Among all the real-time object detectors, our SaccadeNet achieves the best detection performance, which demonstrates the effectiveness of the proposed detection mechanism.

Many real-world applications, such as city scale traffic monitoring and control, requires large scale re-identification. However, previous ReID methods often failed to address two limitations in existing ReID benchmarks, i.e., low spatiotem-poral coverage and sample imbalance. Notwithstanding their demonstrated success in every single benchmark, they have difficulties in generalizing to unseen environments. As a result, these methods are less applicable in a large scale setting due to poor generalization. In seek for a highly generalizable large-scale ReID method, we present an adversarial domain-invariant feature learning framework (ADIN) that explicitly learns to separate identity-related features from challenging variations, where for the first time "free" annotations in ReID data such as video timestamp and camera index are utilized. Furthermore, we find that the imbalance of nuisance classes jeopardizes the adversarial training, and for mitigation we propose a calibrated adversarial loss that is attentive to nuisance distribution. Experiments on existing large-scale person/vehicle ReID datasets demonstrate that ADIN learns more robust and generalizable representations, as evidenced by its outstanding direct transfer performance across datasets, which is a criterion that can better measure the generalizability of large scale Re-ID methods.

This work addresses a novel and challenging problem of estimating the full 3D hand shape and pose from a single RGB image. Most current methods in 3D hand analysis from monocular RGB images only focus on estimating the 3D locations of hand keypoints, which cannot fully express the 3D shape of hand. In contrast, we propose a Graph Convolutional Neural Network (Graph CNN) based method to reconstruct a full 3D mesh of hand surface that contains richer information of both 3D hand shape and pose. To train networks with full supervision, we create a large-scale synthetic dataset containing both ground truth 3D meshes and 3D poses. When fine-tuning the networks on real-world datasets without 3D ground truth, we propose a weakly-supervised approach by leveraging the depth map as a weak supervision in training. Through extensive evaluations on our proposed new datasets and two public datasets, we show that our proposed method can produce accurate and reasonable 3D hand mesh, and can achieve superior 3D hand pose estimation accuracy when compared with state-of-the-art methods.

Dense video captioning is an extremely challenging task since accurate and coherent description of events in a video requires holistic understanding of video contents as well as contextual reasoning of individual events. Most existing approaches handle this problem by first detecting event proposals from a video and then captioning on a subset of the proposals. As a result, the generated sentences are prone to be redundant or inconsistent since they fail to consider temporal dependency between events. To tackle this challenge, we propose a novel dense video captioning framework, which models temporal dependency across events in a video explicitly and leverages visual and linguistic context from prior events for coherent storytelling. This objective is achieved by 1) integrating an event sequence generation network to select a sequence of event proposals adaptively, and 2) feeding the sequence of event proposals to our sequential video captioning network, which is trained by reinforcement learning with two-level rewards - at both event and episode levels - for better context modeling. The proposed technique achieves outstanding performances on ActivityNet Captions dataset in most metrics.

Though CNNs have achieved the state-of-the-art performance on various vision tasks, they are vulnerable to adversarial examples --- crafted by adding human-imperceptible perturbations to clean images. However, most of the existing adversarial attacks only achieve relatively low success rates under the challenging black-box setting, where the attackers have no knowledge of the model structure and parameters. To this end, we propose to improve the transferability of adversarial examples by creating diverse input patterns. Instead of only using the original images to generate adversarial examples, our method applies random transformations to the input images at each iteration. Extensive experiments on ImageNet show that the proposed attack method can generate adversarial examples that transfer much better to different networks than existing baselines. By evaluating our method against top defense solutions and official baselines from NIPS 2017 adversarial competition, the enhanced attack reaches an average success rate of 73.0%, which outperforms the top-1 attack submission in the NIPS competition by a large margin of 6.6%. We hope that our proposed attack strategy can serve as a strong benchmark baseline for evaluating the robustness of networks to adversaries and the effectiveness of different defense methods in the future. Code is available at https://github.com/cihangxie/DI-2-FGSM.

Attention mechanisms have been found effective for person re-identification (Re-ID). However, the learned "attentive'' features are often not naturally uncorrelated or "diverse'', which compromises the retrieval performance based on the Euclidean distance. We advocate the complementary powers of attention and diversity for Re-ID, by proposing an Attentive but Diverse Network (ABD-Net). ABD-Net seamlessly integrates attention modules and diversity regularizations throughout the entire network to learn features that are representative, robust, and more discriminative. Specifically, we introduce a pair of complementary attention modules, focusing on channel aggregation and position awareness, respectively. Then, we plug in a novel orthogonality constraint that efficiently enforces diversity on both hidden activations and weights. Through an extensive set of ablation study, we verify that the attentive and diverse terms each contributes to the performance boosts of ABD-Net. It consistently outperforms existing state-of-the-art methods on there popular person Re-ID benchmarks.

Different from the fully-supervised action detection problem that is dependent on expensive frame-level annotations, weakly supervised action detection (WSAD) only needs video-level annotations, making it more practical for real-world applications. Existing WSAD methods detect action instances by scoring each video segment (a stack of frames) individually. Most of them fail to model the temporal relations among video segments and cannot effectively characterize action instances possessing latent temporal structure. To alleviate this problem in WSAD, we propose the temporal structure mining (TSM) approach. In TSM, each action instance is modeled as a multi-phase process and phase evolving within an action instance, \emph{i.e.}, the temporal structure, is exploited. Meanwhile, the video background is modeled by a background phase, which separates different action instances in an untrimmed video. In this framework, phase filters are used to calculate the confidence scores of the presence of an action's phases in each segment. Since in the WSAD task, frame-level annotations are not available and thus phase filters cannot be trained directly. To tackle the challenge, we treat each segment's phase as a hidden variable. We use segments' confidence scores from each phase filter to construct a table and determine hidden variables, i.e., phases of segments, by a maximal circulant path discovery along the table. Experiments conducted on three benchmark datasets demonstrate the state-of-the-art performance of the proposed TSM.

Video captioning is a challenging task owing to the complexity of understanding the copious visual information in videos and describing it using natural language. Different from previous work that encodes video information using a single flow, in this work, we introduce a novel Sibling Convolutional Encoder (SibNet) for video captioning, which utilizes a two-branch architecture to collaboratively encode videos. The first content branch encodes the visual content information of the video via autoencoder, and the second semantic branch encodes the semantic information by visual-semantic joint embedding. Then both branches are effectively combined with soft-attention mechanism and finally fed into a RNN decoder to generate captions. With our SibNet explicitly capturing both content and semantic information, the proposed method can better represent the rich information in videos. Extensive experiments on YouTube2Text and MSR-VTT datasets validate that the proposed architecture outperforms existing methods by a large margin across different evaluation metrics.

Image captioning is a challenging problem owing to the complexity in understanding the image content and diverse ways of describing it in natural language. Recent advances in deep neural networks have substantially improved the performance of this task. Most state-of-the-art approaches follow an encoder-decoder framework, which generates captions using a sequential recurrent prediction model. However, in this paper, we introduce a novel decision-making framework for image captioning. We utilize a policy network and a value network to collaboratively generate captions. The policy network serves as a local guidance by providing the confidence of predicting the next word according to the current state. Additionally, the value network serves as a global and lookahead guidance by evaluating all possible extensions of the current state. In essence, it adjusts the goal of predicting the correct words towards the goal of generating captions similar to the ground truth captions. We train both networks using an actor-critic reinforcement learning model, with a novel reward defined by visual-semantic embedding. Extensive experiments and analyses on the Microsoft COCO dataset show that the proposed framework outperforms state-of-the-art approaches across different evaluation metrics.

How to jointly represent images and texts is important for tasks involving both modalities. Visual-semantic embedding models have been recently proposed and shown to be effective. The key idea is that by learning a mapping from images into a semantic text space, the algorithm is able to learn a compact and effective joint representation. However, existing approaches simply map each text concept to a single point in the semantic space. Mapping instead to a density distribution provides many interesting advantages, including better capturing uncertainty about each text concept, and enabling better geometric interpretation of concepts such as inclusion, intersection, etc. In this work, we present a novel Gaussian Visual-Semantic Embedding (GVSE) model, which leverages the visual information to model text concepts as Gaussian distributions in semantic space. Experiments in two tasks, image classification and text-based image retrieval on the large scale MIT Places205 dataset, have demonstrated the superiority of our method over existing approaches, with higher accuracy and better robustness.

In this paper, we are interested in enhancing the expressivity and robustness of part-based models for object representation, in the common scenario where the training data are based on 2D images. To this end, we propose scene-domain active part models (SDAPM), which reconstruct and characterize the 3D geometric statistics between object's parts in 3D scene-domain by using 2D training data in the image-domain alone. And on top of this, we explicitly model and handle occlusions in SDAPM. Together with the developed learning and inference algorithms, such a model provides rich object descriptions, including 2D object and parts localization, 3D landmark shape and camera viewpoint, which offers an effective representation to various image understanding tasks, such as object and parts detection, 3D landmark shape and viewpoint estimation from images. Experiments on the above tasks show that SDAPM outperforms previous part-based models, and thus demonstrates the potential of the proposed technique.

Shape decomposition is a fundamental problem for part-based shape representation. We propose the minimum near-convex decomposition (MNCD) to decompose arbitrary shapes into minimum number of "near-convex" parts. The near-convex shape decomposition is formulated as a discrete optimization problem by minimizing the number of nonintersecting cuts. Two perception rules are imposed as constraints into our objective function to improve the visual naturalness of the decomposition. With the degree of near-convexity a user-specified parameter, our decomposition is robust to local distortions and shape deformation. The optimization can be efficiently solved via binary integer linear programming. Both theoretical analysis and experiment results show that our approach outperforms the state-of-the-art results without introducing redundant parts and thus leads to robust shape representation.

The recently developed depth sensors, e.g., the Kinect sensor, have provided new opportunities for human-computer interaction (HCI). Although great progress has been made by leveraging the Kinect sensor, e.g., in human body tracking, face recognition and human action recognition, robust hand gesture recognition remains an open problem. Compared to the entire human body, the hand is a smaller object with more complex articulations and more easily affected by segmentation errors. It is thus a very challenging problem to recognize hand gestures. This paper focuses on building a robust part-based hand gesture recognition system using Kinect sensor. To handle the noisy hand shapes obtained from the Kinect sensor, we propose a novel distance metric, Finger-Earth Mover's Distance (FEMD), to measure the dissimilarity between hand shapes. As it only matches the finger parts while not the whole hand, it can better distinguish the hand gestures of slight differences. The extensive experiments demonstrate that our hand gesture recognition system is accurate (a 93.2% mean accuracy on a challenging 10-gesture dataset), efficient (average 0.0750 s per frame), robust to hand articulations, distortions and orientation or scale changes, and can work in uncontrolled environments (cluttered backgrounds and lighting conditions). The superiority of our system is further demonstrated in two real-life HCI applications.

Shape decomposition is a fundamental problem for part-based shape representation. We propose a novel shape decomposition method called Minimum Near-Convex Decomposition (MNCD), which decomposes 2D and 3D arbitrary shapes into minimum number of “near-convex” parts. With the degree of near-convexity a user specified parameter, our decomposition is robust to large local distortions and shape deformation. The shape decomposition is formulated as a combinatorial optimization problem by minimizing the number of non-intersection cuts. Two major perception rules are also imposed into our scheme to improve the visual naturalness of the decomposition. The global optimal solution of this challenging discrete optimization problem is obtained by a dynamic subgradient-based branch-and-bound search. Both theoretical analysis and experiment results show that our approach outperforms the state-of-the-art results without introducing redundant parts. Finally we also show the superiority of our method in the application of hand gesture recognition.

Hand gesture based Human-Computer-Interaction (HCI) is one of the most natural and intuitive ways to communicate between people and machines, since it closely mimics how human interact with each other. In this demo, we present a hand gesture recognition system with Kinect sensor, which operates robustly in uncontrolled environments and is insensitive to hand variations and distortions. Our system consists of two major modules, namely, hand detection and gesture recognition. Different from traditional vision-based hand gesture recognition methods that use color-markers for hand detection, our system uses both the depth and color information from Kinect sensor to detect the hand shape, which ensures the robustness in cluttered environments. Besides, to guarantee its robustness to input variations or the distortions caused by the low resolution of Kinect sensor, we apply a novel shape distance metric called Finger-Earth Mover's Distance (FEMD) for hand gesture recognition. Consequently, our system operates accurately and efficiently. In this demo, we demonstrate the performance of our system in two real-life applications, arithmetic computation and rock-paper-scissors game.

The recently developed depth sensors, e.g., the Kinect sensor, have provided new opportunities for human-computer interaction (HCI). Although great progress has been made by leveraging the Kinect sensor, e.g. in human body tracking and body gesture recognition, robust hand gesture recognition remains an open problem. Compared to the entire human body, the hand is a smaller object with more complex articulations and more easily affected by segmentation errors. It is thus a very challenging problem to recognize hand gestures. This paper focuses on building a robust hand gesture recognition system using the Kinect sensor. To handle the noisy hand shape obtained from the Kinect sensor, we propose a novel distance metric for hand dissimilarity measure, called Finger-Earth Mover's Distance (FEMD). As it only matches fingers while not the whole hand shape, it can better distinguish hand gestures of slight differences. The extensive experiments demonstrate the accuracy, efficiency, and robustness of our hand gesture recognition system.

We present two edge encoding schemes, namely 8-sector scheme and 16-sector scheme, for the operational rate-distortion (ORD) optimal shape coding framework. Different from the traditional 8-direction scheme that can only encode edges with angles being an integer multiple of π/4, our proposals can encode edges with arbitrary angles. We partition the digital coordinate plane into 8 and 16 sectors, and design the corresponding differential schemes to encode the short and the long component of each vertex. Experiment results demonstrate that our two proposals can reduce a large number of encoding vertices and therefore reduce 10%~20% bits for the basic ORD optimal algorithms and 10%~30% bits for all the ORD optimal algorithms under the same distortion thresholds, respectively. Moreover, the reconstruction contours are more compact compared with those using the traditional 8-direction edge encoding scheme.