There are multiple cues in an image which reveal what action a person is performing. For example, a jogger has a pose that is characteristic for jogging, but the scene (e.g. road, trail) and the presence of other joggers can be an additional source of information. In this work, we exploit the simple observation that actions are accompanied by contextual cues to build a strong action recognition system. We adapt RCNN to use more than one region for classification while still maintaining the ability to localize the action. We call our system R*CNN. The action-specific models and the feature maps are trained jointly, allowing for action specific representations to emerge. R*CNN achieves 90.2% mean AP on the PASAL VOC Action dataset, outperforming all other approaches in the field by a significant margin. Last, we show that R*CNN is not limited to action recognition. In particular, R*CNN can also be used to tackle fine-grained tasks such as attribute classification. We validate this claim by reporting state-of-the-art performance on the Berkeley Attributes of People dataset.
This paper presents a novel method for learning a pose lexicon comprising semantic poses defined by textual instructions and their associated visual poses defined by visual features. The proposed method simultaneously takes two input streams, semantic poses and visual pose candidates, and statistically learns a mapping between them to construct the lexicon. With the learned lexicon, action recognition can be cast as the problem of finding the maximum translation probability of a sequence of semantic poses given a stream of visual pose candidates. Experiments evaluating pre-trained and zero-shot action recognition conducted on MSRC-12 gesture and WorkoutSu-10 exercise datasets were used to verify the efficacy of the proposed method.
Egocentric videos are characterised by their ability to have the first person view. With the popularity of Google Glass and GoPro, use of egocentric videos is on the rise. Recognizing action of the wearer from egocentric videos is an important problem. Unstructured movement of the camera due to natural head motion of the wearer causes sharp changes in the visual field of the egocentric camera causing many standard third person action recognition techniques to perform poorly on such videos. Objects present in the scene and hand gestures of the wearer are the most important cues for first person action recognition but are difficult to segment and recognize in an egocentric video. We propose a novel representation of the first person actions derived from feature trajectories. The features are simple to compute using standard point tracking and does not assume segmentation of hand/objects or recognizing object or hand pose unlike in many previous approaches. We train a bag of words classifier with the proposed features and report a performance improvement of more than 11% on publicly available datasets. Although not designed for the particular case, we show that our technique can also recognize wearer's actions when hands or objects are not visible.
How can unlabeled video augment visual learning? Existing methods perform "slow" feature analysis, encouraging the representations of temporally close frames to exhibit only small differences. While this standard approach captures the fact that high-level visual signals change slowly over time, it fails to capture *how* the visual content changes. We propose to generalize slow feature analysis to "steady" feature analysis. The key idea is to impose a prior that higher order derivatives in the learned feature space must be small. To this end, we train a convolutional neural network with a regularizer on tuples of sequential frames from unlabeled video. It encourages feature changes over time to be smooth, i.e., similar to the most recent changes. Using five diverse datasets, including unlabeled YouTube and KITTI videos, we demonstrate our method's impact on object, scene, and action recognition tasks. We further show that our features learned from unlabeled video can even surpass a standard heavily supervised pretraining approach.
Automatically recognizing and localizing wide ranges of human actions has crucial importance for video understanding. Towards this goal, the THUMOS challenge was introduced in 2013 to serve as a benchmark for action recognition. Until then, video action recognition, including THUMOS challenge, had focused primarily on the classification of pre-segmented (i.e., trimmed) videos, which is an artificial task. In THUMOS 2014, we elevated action recognition to a more practical level by introducing temporally untrimmed videos. These also include `background videos' which share similar scenes and backgrounds as action videos, but are devoid of the specific actions. The three editions of the challenge organized in 2013--2015 have made THUMOS a common benchmark for action classification and detection and the annual challenge is widely attended by teams from around the world. In this paper we describe the THUMOS benchmark in detail and give an overview of data collection and annotation procedures. We present the evaluation protocols used to quantify results in the two THUMOS tasks of action classification and temporal detection. We also present results of submissions to the THUMOS 2015 challenge and review the participating approaches. Additionally, we include a comprehensive empirical study evaluating the differences in action recognition between trimmed and untrimmed videos, and how well methods trained on trimmed videos generalize to untrimmed videos. We conclude by proposing several directions and improvements for future THUMOS challenges.
In this paper we consider the task of recognizing human actions in realistic video where human actions are dominated by irrelevant factors. We first study the benefits of removing non-action video segments, which are the ones that do not portray any human action. We then learn a non-action classifier and use it to down-weight irrelevant video segments. The non-action classifier is trained using ActionThread, a dataset with shot-level annotation for the occurrence or absence of a human action. The non-action classifier can be used to identify non-action shots with high precision and subsequently used to improve the performance of action recognition systems.
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.
The deep two-stream architecture exhibited excellent performance on video based action recognition. The most computationally expensive step in this approach comes from the calculation of optical flow which prevents it to be real-time. This paper accelerates this architecture by replacing optical flow with motion vector which can be obtained directly from compressed videos without extra calculation. However, motion vector lacks fine structures, and contains noisy and inaccurate motion patterns, leading to the evident degradation of recognition performance. Our key insight for relieving this problem is that optical flow and motion vector are inherent correlated. Transferring the knowledge learned with optical flow CNN to motion vector CNN can significantly boost the performance of the latter. Specifically, we introduce three strategies for this, initialization transfer, supervision transfer and their combination. Experimental results show that our method achieves comparable recognition performance to the state-of-the-art, while our method can process 390.7 frames per second, which is 27 times faster than the original two-stream method.
Actionness was introduced to quantify the likelihood of containing a generic action instance at a specific location. Accurate and efficient estimation of actionness is important in video analysis and may benefit other relevant tasks such as action recognition and action detection. This paper presents a new deep architecture for actionness estimation, called hybrid fully convolutional network (H-FCN), which is composed of appearance FCN (A-FCN) and motion FCN (M-FCN). These two FCNs leverage the strong capacity of deep models to estimate actionness maps from the perspectives of static appearance and dynamic motion, respectively. In addition, the fully convolutional nature of H-FCN allows it to efficiently process videos with arbitrary sizes. Experiments are conducted on the challenging datasets of Stanford40, UCF Sports, and JHMDB to verify the effectiveness of H-FCN on actionness estimation, which demonstrate that our method achieves superior performance to previous ones. Moreover, we apply the estimated actionness maps on action proposal generation and action detection. Our actionness maps advance the current state-of-the-art performance of these tasks substantially.
By thoroughly revisiting the classic human action recognition paradigm, this paper aims at proposing a new approach for the design of effective action classification systems. Taking as testbed publicly available three-dimensional (MoCap) action/activity datasets, we analyzed and validated different training/testing strategies. In particular, considering that each human action in the datasets is performed several times by different subjects, we were able to precisely quantify the effect of inter- and intra-subject variability, so as to figure out the impact of several learning approaches in terms of classification performance. The net result is that standard testing strategies consisting in cross-validating the algorithm using typical splits of the data (holdout, k-fold, or one-subject-out) is always outperformed by a "personalization" strategy which learns how a subject is performing an action. In other words, it is advantageous to customize (i.e., personalize) the method to learn the actions carried out by each subject, rather than trying to generalize the actions executions across subjects. Consequently, we finally propose an action recognition framework consisting of a two-stage classification approach where, given a test action, the subject is first identified before the actual recognition of the action takes place. Despite the basic, off-the-shelf descriptors and standard classifiers adopted, we noted a relevant increase in performance with respect to standard state-of-the-art algorithms, so motivating the usage of personalized approaches for designing effective action recognition systems.
In this paper, we propose a framework for recognizing human activities that uses only in-topic dominant codewords and a mixture of intertopic vectors. Latent Dirichlet allocation (LDA) is used to develop approximations of human motion primitives; these are mid-level representations, and they adaptively integrate dominant vectors when classifying human activities. In LDA topic modeling, action videos (documents) are represented by a bag-of-words (input from a dictionary), and these are based on improved dense trajectories. The output topics correspond to human motion primitives, such as finger moving or subtle leg motion. We eliminate the impurities, such as missed tracking or changing light conditions, in each motion primitive. The assembled vector of motion primitives is an improved representation of the action. We demonstrate our method on four different datasets.
Improved dense trajectories (iDT) have shown great performance in action recognition, and their combination with the two-stream approach has achieved state-of-the-art performance. It is, however, difficult for iDT to completely remove background trajectories from video with camera shaking. Trajectories in less discriminative regions should be given modest weights in order to create more discriminative local descriptors for action recognition. In addition, the two-stream approach, which learns appearance and motion information separately, cannot focus on motion in important regions when extracting features from spatial convolutional layers of the appearance network, and vice versa. In order to address the above mentioned problems, we propose a new local descriptor that pools a new convolutional layer obtained from crossing two networks along iDT. This new descriptor is calculated by applying discriminative weights learned from one network to a convolutional layer of the other network. Our method has achieved state-of-the-art performance on ordinal action recognition datasets, 92.3% on UCF101, and 66.2% on HMDB51.
This work presents an approach to category-based action recognition in video using sparse coding techniques. The proposed approach includes two main contributions: i) A new method to handle intra-class variations by decomposing each video into a reduced set of representative atomic action acts or key-sequences, and ii) A new video descriptor, ITRA: Inter-Temporal Relational Act Descriptor, that exploits the power of comparative reasoning to capture relative similarity relations among key-sequences. In terms of the method to obtain key-sequences, we introduce a loss function that, for each video, leads to the identification of a sparse set of representative key-frames capturing both, relevant particularities arising in the input video, as well as relevant generalities arising in the complete class collection. In terms of the method to obtain the ITRA descriptor, we introduce a novel scheme to quantify relative intra and inter-class similarities among local temporal patterns arising in the videos. The resulting ITRA descriptor demonstrates to be highly effective to discriminate among action categories. As a result, the proposed approach reaches remarkable action recognition performance on several popular benchmark datasets, outperforming alternative state-of-the-art techniques by a large margin.
We bring together ideas from recent work on feature design for egocentric action recognition under one framework by exploring the use of deep convolutional neural networks (CNN). Recent work has shown that features such as hand appearance, object attributes, local hand motion and camera ego-motion are important for characterizing first-person actions. To integrate these ideas under one framework, we propose a twin stream network architecture, where one stream analyzes appearance information and the other stream analyzes motion information. Our appearance stream encodes prior knowledge of the egocentric paradigm by explicitly training the network to segment hands and localize objects. By visualizing certain neuron activation of our network, we show that our proposed architecture naturally learns features that capture object attributes and hand-object configurations. Our extensive experiments on benchmark egocentric action datasets show that our deep architecture enables recognition rates that significantly outperform state-of-the-art techniques -- an average $6.6\%$ increase in accuracy over all datasets. Furthermore, by learning to recognize objects, actions and activities jointly, the performance of individual recognition tasks also increase by $30\%$ (actions) and $14\%$ (objects). We also include the results of extensive ablative analysis to highlight the importance of network design decisions..
We propose a function-based temporal pooling method that captures the latent structure of the video sequence data - e.g. how frame-level features evolve over time in a video. We show how the parameters of a function that has been fit to the video data can serve as a robust new video representation. As a specific example, we learn a pooling function via ranking machines. By learning to rank the frame-level features of a video in chronological order, we obtain a new representation that captures the video-wide temporal dynamics of a video, suitable for action recognition. Other than ranking functions, we explore different parametric models that could also explain the temporal changes in videos. The proposed functional pooling methods, and rank pooling in particular, is easy to interpret and implement, fast to compute and effective in recognizing a wide variety of actions. We evaluate our method on various benchmarks for generic action, fine-grained action and gesture recognition. Results show that rank pooling brings an absolute improvement of 7-10 average pooling baseline. At the same time, rank pooling is compatible with and complementary to several appearance and local motion based methods and features, such as improved trajectories and deep learning features.
Detecting and recognizing objects interacting with humans lie in the center of first-person (egocentric) daily activity recognition. However, due to noisy camera motion and frequent changes in viewpoint and scale, most of the previous egocentric action recognition methods fail to capture and model highly discriminative object features. In this work, we propose a novel pipeline for first-person daily activity recognition, aiming at more discriminative object feature representation and object-motion feature fusion. Our object feature extraction and representation pipeline is inspired by the recent success of object hypotheses and deep convolutional neural network based detection frameworks. Our key contribution is a simple yet effective manipulated object proposal generation scheme. This scheme leverages motion cues such as motion boundary and motion magnitude (in contrast, camera motion is usually considered as "noise" for most previous methods) to generate a more compact and discriminative set of object proposals, which are more closely related to the objects which are being manipulated. Then, we learn more discriminative object detectors from these manipulated object proposals based on region-based convolutional neural network (R-CNN). Meanwhile, we develop a network based feature fusion scheme which better combines object and motion features. We show in experiments that the proposed framework significantly outperforms the state-of-the-art recognition performance on a challenging first-person daily activity benchmark.
Detecting hand actions from ego-centric depth sequences is a practically challenging problem, owing mostly to the complex and dexterous nature of hand articulations as well as non-stationary camera motion. We address this problem via a Hough transform based approach coupled with a discriminatively learned error-correcting component to tackle the well known issue of incorrect votes from the Hough transform. In this framework, local parts vote collectively for the start $\&$ end positions of each action over time. We also construct an in-house annotated dataset of 300 long videos, containing 3,177 single-action subsequences over 16 action classes collected from 26 individuals. Our system is empirically evaluated on this real-life dataset for both the action recognition and detection tasks, and is shown to produce satisfactory results. To facilitate reproduction, the new dataset and our implementation are also provided online.
In this paper, we introduce a new hierarchical model for human action recognition using body joint locations. Our model can categorize complex actions in videos, and perform spatio-temporal annotations of the atomic actions that compose the complex action being performed.That is, for each atomic action, the model generates temporal action annotations by estimating its starting and ending times, as well as, spatial annotations by inferring the human body parts that are involved in executing the action. our model includes three key novel properties: (i) it can be trained with no spatial supervision, as it can automatically discover active body parts from temporal action annotations only; (ii) it jointly learns flexible representations for motion poselets and actionlets that encode the visual variability of body parts and atomic actions; (iii) a mechanism to discard idle or non-informative body parts which increases its robustness to common pose estimation errors. We evaluate the performance of our method using multiple action recognition benchmarks. Our model consistently outperforms baselines and state-of-the-art action recognition methods.
In recent years significant progress has been made in successfully training recurrent neural networks (RNNs) on sequence learning problems involving long range temporal dependencies. The progress has been made on three fronts: (a) Algorithmic improvements involving sophisticated optimization techniques, (b) network design involving complex hidden layer nodes and specialized recurrent layer connections and (c) weight initialization methods. In this paper, we focus on recently proposed weight initialization with identity matrix for the recurrent weights in a RNN. This initialization is specifically proposed for hidden nodes with Rectified Linear Unit (ReLU) non linearity. We offer a simple dynamical systems perspective on weight initialization process, which allows us to propose a modified weight initialization strategy. We show that this initialization technique leads to successfully training RNNs composed of ReLUs. We demonstrate that our proposal produces comparable or better solution for three toy problems involving long range temporal structure: the addition problem, the multiplication problem and the MNIST classification problem using sequence of pixels. In addition, we present results for a benchmark action recognition problem.
Annotated datasets are commonly used in the training and evaluation of tasks involving natural language and vision (image description generation, action recognition and visual question answering). However, many of the existing datasets reflect problems that emerge in the process of data selection and annotation. Here we point out some of the difficulties and problems one confronts when creating and validating annotated vision and language datasets.