We introduce OmniSource, a novel framework for leveraging web data to train video recognition models. OmniSource overcomes the barriers between data formats, such as images, short videos, and long untrimmed videos for webly-supervised learning. First, data samples with multiple formats, curated by task-specific data collection and automatically filtered by a teacher model, are transformed into a unified form. Then a joint-training strategy is proposed to deal with the domain gaps between multiple data sources and formats in webly-supervised learning. Several good practices, including data balancing, resampling, and cross-dataset mixup are adopted in joint training. Experiments show that by utilizing data from multiple sources and formats, OmniSource is more data-efficient in training. With only 3.5M images and 800K minutes videos crawled from the internet without human labeling (less than 2% of prior works), our models learned with OmniSource improve Top-1 accuracy of 2D- and 3D-ConvNet baseline models by 3.0% and 3.9%, respectively, on the Kinetics-400 benchmark. With OmniSource, we establish new records with different pretraining strategies for video recognition. Our best models achieve 80.4%, 80.5%, and 83.6 Top-1 accuracies on the Kinetics-400 benchmark respectively for training-from-scratch, ImageNet pre-training and IG-65M pre-training.
Despite the fact that notable improvements have been made recently in the field of feature extraction and classification, human action recognition is still challenging, especially in images, in which, unlike videos, there is no motion. Thus, the methods proposed for recognizing human actions in videos cannot be applied to still images. A big challenge in action recognition in still images is the lack of large enough datasets, which is problematic for training deep Convolutional Neural Networks (CNNs) due to the overfitting issue. In this paper, by taking advantage of pre-trained CNNs, we employ the transfer learning technique to tackle the lack of massive labeled action recognition datasets. Furthermore, since the last layer of the CNN has class-specific information, we apply an attention mechanism on the output feature maps of the CNN to extract more discriminative and powerful features for classification of human actions. Moreover, we use eight different pre-trained CNNs in our framework and investigate their performance on Stanford 40 dataset. Finally, we propose using the Ensemble Learning technique to enhance the overall accuracy of action classification by combining the predictions of multiple models. The best setting of our method is able to achieve 93.17$\%$ accuracy on the Stanford 40 dataset.
Zero-shot learning strives to classify unseen categories for which no data is available during training. In the generalized variant, the test samples can further belong to seen or unseen categories. The state-of-the-art relies on Generative Adversarial Networks that synthesize unseen class features by leveraging class-specific semantic embeddings. During training, they generate semantically consistent features, but discard this constraint during feature synthesis and classification. We propose to enforce semantic consistency at all stages of (generalized) zero-shot learning: training, feature synthesis and classification. We first introduce a feedback loop, from a semantic embedding decoder, that iteratively refines the generated features during both the training and feature synthesis stages. The synthesized features together with their corresponding latent embeddings from the decoder are then transformed into discriminative features and utilized during classification to reduce ambiguities among categories. Experiments on (generalized) zero-shot object and action classification reveal the benefit of semantic consistency and iterative feedback, outperforming existing methods on six zero-shot learning benchmarks. Source code at https://github.com/akshitac8/tfvaegan.
A challenge of skeleton-based action recognition is the difficulty to classify actions with similar motions and object-related actions. Visual clues from other streams help in that regard. RGB data are sensible to illumination conditions, thus unusable in the dark. To alleviate this issue and still benefit from a visual stream, we propose a modular network (FUSION) combining skeleton and infrared data. A 2D convolutional neural network (CNN) is used as a pose module to extract features from skeleton data. A 3D CNN is used as an infrared module to extract visual cues from videos. Both feature vectors are then concatenated and exploited conjointly using a multilayer perceptron (MLP). Skeleton data also condition the infrared videos, providing a crop around the performing subjects and thus virtually focusing the attention of the infrared module. Ablation studies show that using pre-trained networks on other large scale datasets as our modules and data augmentation yield considerable improvements on the action classification accuracy. The strong contribution of our cropping strategy is also demonstrated. We evaluate our method on the NTU RGB+D dataset, the largest dataset for human action recognition from depth cameras, and report state-of-the-art performances.
This paper proposes a simple yet effective method for human action recognition in video. The proposed method separately extracts local appearance and motion features using state-of-the-art three-dimensional convolutional neural networks from sampled snippets of a video. These local features are then concatenated to form global representations which are then used to train a linear SVM to perform the action classification using full context of the video, as partial context as used in previous works. The videos undergo two simple proposed preprocessing techniques, optical flow scaling and crop filling. We perform an extensive evaluation on three common benchmark dataset to empirically show the benefit of the SVM, and the two preprocessing steps.
Deep neural networks for video classification, just like image classification networks, may be subjected to adversarial manipulation. The main difference between image classifiers and video classifiers is that the latter usually use temporal information contained within the video. In this work we present a manipulation scheme for fooling video classifiers by introducing a flickering temporal perturbation that in some cases may be unnoticeable by human observers and is implementable in the real world. After demonstrating the manipulation of action classification of single videos, we generalize the procedure to make universal adversarial perturbation, achieving high fooling ratio. In addition, we generalize the universal perturbation and produce a temporal-invariant perturbation, which can be applied to the video without synchronizing the perturbation to the input. The attack was implemented on several target models and the transferability of the attack was demonstrated. These properties allow us to bridge the gap between simulated environment and real-world application, as will be demonstrated in this paper for the first time for an over-the-air flickering attack.
We present Audiovisual SlowFast Networks, an architecture for integrated audiovisual perception. AVSlowFast has Slow and Fast visual pathways that are deeply integrated with a Faster Audio pathway to model vision and sound in a unified representation. We fuse audio and visual features at multiple layers, enabling audio to contribute to the formation of hierarchical audiovisual concepts. To overcome training difficulties that arise from different learning dynamics for audio and visual modalities, we introduce DropPathway, which randomly drops the Audio pathway during training as an effective regularization technique. Inspired by prior studies in neuroscience, we perform hierarchical audiovisual synchronization to learn joint audiovisual features. We report state-of-the-art results on six video action classification and detection datasets, perform detailed ablation studies, and show the generalization of AVSlowFast to learn self-supervised audiovisual features. Code will be made available at: https://github.com/facebookresearch/SlowFast.
Current video representations heavily rely on learning from manually annotated video datasets which are time-consuming and expensive to acquire. We observe videos are naturally accompanied by abundant text information such as YouTube titles and Instagram captions. In this paper, we leverage this visual-textual connection to learn spatiotemporal features in an efficient weakly-supervised manner. We present a general cross-modal pair discrimination (CPD) framework to capture this correlation between a video and its associated text. Specifically, we adopt noise-contrastive estimation to tackle the computational issue imposed by the huge amount of pair instance classes and design a practical curriculum learning strategy. We train our CPD models on both standard video dataset (Kinetics-210k) and uncurated web video dataset (Instagram-300k) to demonstrate its effectiveness. Without further fine-tuning, the learnt models obtain competitive results for action classification on Kinetics under the linear classification protocol. Moreover, our visual model provides an effective initialization to fine-tune on downstream tasks, which yields a remarkable performance gain for action recognition on UCF101 and HMDB51, compared with the existing state-of-the-art self-supervised training methods. In addition, our CPD model yields a new state of the art for zero-shot action recognition on UCF101 by directly utilizing the learnt visual-textual embeddings. The code will be made available at https://github.com/MCG-NJU/CPD-Video.
In this paper, we build on a concept of self-supervision by taking RGB frames as input to learn to predict both action concepts and auxiliary descriptors e.g., object descriptors. So-called hallucination streams are trained to predict auxiliary cues, simultaneously fed into classification layers, and then hallucinated at the testing stage to aid network. We design and hallucinate two descriptors, one leveraging four popular object detectors applied to training videos, and the other leveraging image- and video-level saliency detectors. The first descriptor encodes the detector- and ImageNet-wise class prediction scores, confidence scores, and spatial locations of bounding boxes and frame indexes to capture the spatio-temporal distribution of features per video. Another descriptor encodes spatio-angular gradient distributions of saliency maps and intensity patterns. Inspired by the characteristic function of the probability distribution, we capture four statistical moments on the above intermediate descriptors. As numbers of coefficients in the mean, covariance, coskewness and cokurtotsis grow linearly, quadratically, cubically and quartically w.r.t. the dimension of feature vectors, we describe the covariance matrix by its leading n' eigenvectors (so-called subspace) and we capture skewness/kurtosis rather than costly coskewness/cokurtosis. We obtain state of the art on five popular datasets such as Charades and EPIC-Kitchens.
Graph and language embedding models are becoming commonplace in large scale analyses given their ability to represent complex sparse data densely in low-dimensional space. Integrating these models' complementary relational and communicative data may be especially helpful if predicting rare events or classifying members of hidden populations - tasks requiring huge and sparse datasets for generalizable analyses. For example, due to social stigma and comorbidities, mental health support groups often form in amorphous online groups. Predicting suicidality among individuals in these settings using standard network analyses is prohibitive due to resource limits (e.g., memory), and adding auxiliary data like text to such models exacerbates complexity- and sparsity-related issues. Here, I show how merging graph and language embedding models (metapath2vec and doc2vec) avoids these limits and extracts unsupervised clustering data without domain expertise or feature engineering. Graph and language distances to a suicide support group have little correlation (\r{ho} < 0.23), implying the two models are not embedding redundant information. When used separately to predict suicidality among individuals, graph and language data generate relatively accurate results (69% and 76%, respectively); however, when integrated, both data produce highly accurate predictions (90%, with 10% false-positives and 12% false-negatives). Visualizing graph embeddings annotated with predictions of potentially suicidal individuals shows the integrated model could classify such individuals even if they are positioned far from the support group. These results extend research on the importance of simultaneously analyzing behavior and language in massive networks and efforts to integrate embedding models for different kinds of data when predicting and classifying, particularly when they involve rare events.
We present a data-driven deep neural algorithm for detecting deceptive walking behavior using nonverbal cues like gaits and gestures. We conducted an elaborate user study, where we recorded many participants performing tasks involving deceptive walking. We extract the participants' walking gaits as series of 3D poses. We annotate various gestures performed by participants during their tasks. Based on the gait and gesture data, we train an LSTM-based deep neural network to obtain deep features. Finally, we use a combination of psychology-based gait, gesture, and deep features to detect deceptive walking with an accuracy of 88.41%. This is an improvement of 10.6% over handcrafted gait and gesture features and an improvement of 4.7% and 9.2% over classifiers based on the state-of-the-art emotion and action classification algorithms, respectively. Additionally, we present a novel dataset, DeceptiveWalk, that contains gaits and gestures with their associated deception labels. To the best of our knowledge, ours is the first algorithm to detect deceptive behavior using non-verbal cues of gait and gesture.
Soccer videos can serve as a perfect research object for video understanding because soccer games are played under well-defined rules while complex and intriguing enough for researchers to study. In this paper, we propose a new soccer video database named SoccerDB, comprising 171,191 video segments from 346 high-quality soccer games. The database contains 702,096 bounding boxes, 37,709 essential event labels with time boundary and 17,115 highlight annotations for object detection, action recognition, temporal action localization, and highlight detection tasks. To our knowledge, it is the largest database for comprehensive sports video understanding on various aspects. We further survey a collection of strong baselines on SoccerDB, which have demonstrated state-of-the-art performances on independent tasks. Our evaluation suggests that we can benefit significantly when jointly considering the inner correlations among those tasks. We believe the release of SoccerDB will tremendously advance researches around comprehensive video understanding. {\itshape Our dataset and code published on https://github.com/newsdata/SoccerDB.}
Spatiotemporal representations learned using 3D convolutional neural networks (CNN) are currently used in state-of-the-art approaches for action related tasks. However, 3D-CNN are notorious for being memory and compute resource intensive as compared with more simple 2D-CNN architectures. We propose to hallucinate spatiotemporal representations from a 3D-CNN teacher with a 2D-CNN student. By requiring the 2D-CNN to predict the future and intuit upcoming activity, it is encouraged to gain a deeper understanding of actions and how they evolve. The hallucination task is treated as an auxiliary task, which can be used with any other action related task in a multitask learning setting. Thorough experimental evaluation shows that the hallucination task indeed helps improve performance on action recognition, action quality assessment, and dynamic scene recognition tasks. From a practical standpoint, being able to hallucinate spatiotemporal representations without an actual 3D-CNN can enable deployment in resource-constrained scenarios, such as with limited computing power and/or lower bandwidth. Codebase is available here: https://github.com/ParitoshParmar/HalluciNet.
Human activities often occur in specific scene contexts, e.g., playing basketball on a basketball court. Training a model using existing video datasets thus inevitably captures and leverages such bias (instead of using the actual discriminative cues). The learned representation may not generalize well to new action classes or different tasks. In this paper, we propose to mitigate scene bias for video representation learning. Specifically, we augment the standard cross-entropy loss for action classification with 1) an adversarial loss for scene types and 2) a human mask confusion loss for videos where the human actors are masked out. These two losses encourage learning representations that are unable to predict the scene types and the correct actions when there is no evidence. We validate the effectiveness of our method by transferring our pre-trained model to three different tasks, including action classification, temporal localization, and spatio-temporal action detection. Our results show consistent improvement over the baseline model without debiasing.
Pre-training convolutional neural networks with weakly-supervised and self-supervised strategies is becoming increasingly popular for several computer vision tasks. However, due to the lack of strong discriminative signals, these learned representations may overfit to the pre-training objective (e.g., hashtag prediction) and not generalize well to downstream tasks. In this work, we present a simple strategy - ClusterFit (CF) to improve the robustness of the visual representations learned during pre-training. Given a dataset, we (a) cluster its features extracted from a pre-trained network using k-means and (b) re-train a new network from scratch on this dataset using cluster assignments as pseudo-labels. We empirically show that clustering helps reduce the pre-training task-specific information from the extracted features thereby minimizing overfitting to the same. Our approach is extensible to different pre-training frameworks -- weak- and self-supervised, modalities -- images and videos, and pre-training tasks -- object and action classification. Through extensive transfer learning experiments on 11 different target datasets of varied vocabularies and granularities, we show that ClusterFit significantly improves the representation quality compared to the state-of-the-art large-scale (millions / billions) weakly-supervised image and video models and self-supervised image models.
Although synthetic training data has been shown to be beneficial for tasks such as human pose estimation, its use for RGB human action recognition is relatively unexplored. Our goal in this work is to answer the question whether synthetic humans can improve the performance of human action recognition, with a particular focus on generalization to unseen viewpoints. We make use of the recent advances in monocular 3D human body reconstruction from real action sequences to automatically render synthetic training videos for the action labels. We make the following contributions: (i) we investigate the extent of variations and augmentations that are beneficial to improving performance at new viewpoints. We consider changes in body shape and clothing for individuals, as well as more action relevant augmentations such as non-uniform frame sampling, and interpolating between the motion of individuals performing the same action; (ii) We introduce a new data generation methodology, SURREACT, that allows training of spatio-temporal CNNs for action classification; (iii) We substantially improve the state-of-the-art action recognition performance on the NTU RGB+D and UESTC standard human action multi-view benchmarks; Finally, (iv) we extend the augmentation approach to in-the-wild videos from a subset of the Kinetics dataset to investigate the case when only one-shot training data is available, and demonstrate improvements in this case as well.
Current state-of-the-art models for video action recognition are mostly based on expensive 3D ConvNets. This results in a need for large GPU clusters to train and evaluate such architectures. To address this problem, we present a lightweight and memory-friendly architecture for action recognition that performs on par with or better than current architectures by using only a fraction of resources. The proposed architecture is based on a combination of a deep subnet operating on low-resolution frames with a compact subnet operating on high-resolution frames, allowing for high efficiency and accuracy at the same time. We demonstrate that our approach achieves a reduction by $3\sim4$ times in FLOPs and $\sim2$ times in memory usage compared to the baseline. This enables training deeper models with more input frames under the same computational budget. To further obviate the need for large-scale 3D convolutions, a temporal aggregation module is proposed to model temporal dependencies in a video at very small additional computational costs. Our models achieve strong performance on several action recognition benchmarks including Kinetics, Something-Something and Moments-in-time. The code and models are available at https://github.com/IBM/bLVNet-TAM.
The point process is a solid framework to model sequential data, such as videos, by exploring the underlying relevance. As a challenging problem for high-level video understanding, weakly supervised action recognition and localization in untrimmed videos has attracted intensive research attention. Knowledge transfer by leveraging the publicly available trimmed videos as external guidance is a promising attempt to make up for the coarse-grained video-level annotation and improve the generalization performance. However, unconstrained knowledge transfer may bring about irrelevant noise and jeopardize the learning model. This paper proposes a novel adaptability decomposing encoder-decoder network to transfer reliable knowledge between trimmed and untrimmed videos for action recognition and localization via bidirectional point process modeling, given only video-level annotations. By decomposing the original features into domain-adaptable and domain-specific ones based on their adaptability, trimmed-untrimmed knowledge transfer can be safely confined within a more coherent subspace. An encoder-decoder based structure is carefully designed and jointly optimized to facilitate effective action classification and temporal localization. Extensive experiments are conducted on two benchmark datasets (i.e., THUMOS14 and ActivityNet1.3), and experimental results clearly corroborate the efficacy of our method.
In recent years, surveillance cameras are widely deployed in public places, and the general crime rate has been reduced significantly due to these ubiquitous devices. Usually, these cameras provide cues and evidence after crimes are conducted, while they are rarely used to prevent or stop criminal activities in time. It is both time and labor consuming to manually monitor a large amount of video data from surveillance cameras. Therefore, automatically recognizing violent behaviors from video signals becomes essential. This paper summarizes several existing video datasets for violence detection and proposes the RWF-2000 database with 2,000 videos captured by surveillance cameras in real-world scenes. Also, we present a new method that utilizes both the merits of 3D-CNNs and optical flow, namely Flow Gated Network. The proposed approach obtains an accuracy of 87.25% on the test set of our proposed database. The database and source codes are currently open to access.
Vision-based sign language recognition aims at helping deaf people to communicate with others. However, most existing sign language datasets are limited to a small number of words. Due to the limited vocabulary size, models learned from those datasets cannot be applied in practice. In this paper, we introduce a new large-scale Word-Level American Sign Language (WLASL) video dataset, containing more than 2000 words performed by over 100 signers. This dataset will be made publicly available to the research community. To our knowledge, it is by far the largest public ASL dataset to facilitate word-level sign recognition research. Based on this new large-scale dataset, we are able to experiment with several deep learning methods for word-level sign recognition and evaluate their performances in large scale scenarios. Specifically we implement and compare two different models,i.e., (i) holistic visual appearance-based approach, and (ii) 2D human pose based approach. Both models are valuable baselines that will benefit the community for method benchmarking. Moreover, we also propose a novel pose-based temporal graph convolution networks (Pose-TGCN) that models spatial and temporal dependencies in human pose trajectories simultaneously, which has further boosted the performance of the pose-based method. Our results show that pose-based and appearance-based models achieve comparable performances up to 66% at top-10 accuracy on 2,000 words/glosses, demonstrating the validity and challenges of our dataset. Our dataset and baseline deep models are available at \url{https://dxli94.github.io/WLASL/}.