We present a convolution-free approach to video classification built exclusively on self-attention over space and time. Our method, named "TimeSformer," adapts the standard Transformer architecture to video by enabling spatiotemporal feature learning directly from a sequence of frame-level patches. Our experimental study compares different self-attention schemes and suggests that "divided attention," where temporal attention and spatial attention are separately applied within each block, leads to the best video classification accuracy among the design choices considered. Despite the radically new design, TimeSformer achieves state-of-the-art results on several action recognition benchmarks, including the best reported accuracy on Kinetics-400 and Kinetics-600. Finally, compared to 3D convolutional networks, our model is faster to train, it can achieve dramatically higher test efficiency (at a small drop in accuracy), and it can also be applied to much longer video clips (over one minute long). Code and models are available at: https://github.com/facebookresearch/TimeSformer.
This paper presents VTN, a transformer-based framework for video recognition. Inspired by recent developments in vision transformers, we ditch the standard approach in video action recognition that relies on 3D ConvNets and introduce a method that classifies actions by attending to the entire video sequence information. Our approach is generic and builds on top of any given 2D spatial network. In terms of wall runtime, it trains $16.1\times$ faster and runs $5.1\times$ faster during inference while maintaining competitive accuracy compared to other state-of-the-art methods. It enables whole video analysis, via a single end-to-end pass, while requiring $1.5\times$ fewer GFLOPs. We report competitive results on Kinetics-400 and present an ablation study of VTN properties and the trade-off between accuracy and inference speed. We hope our approach will serve as a new baseline and start a fresh line of research in the video recognition domain. Code and models are available at: https://github.com/bomri/SlowFast/blob/master/projects/vtn/README.md
Contrastive learning has nearly closed the gap between supervised and self-supervised learning of image representations, and has also been explored for videos. However, prior work on contrastive learning for video data has not explored the effect of explicitly encouraging the features to be distinct across the temporal dimension. We develop a new temporal contrastive learning framework consisting of two novel losses to improve upon existing contrastive self-supervised video representation learning methods. The local-local temporal contrastive loss adds the task of discriminating between non-overlapping clips from the same video, whereas the global-local temporal contrastive aims to discriminate between timesteps of the feature map of an input clip in order to increase the temporal diversity of the learned features. Our proposed temporal contrastive learning framework achieves significant improvement over the state-of-the-art results in various downstream video understanding tasks such as action recognition, limited-label action classification, and nearest-neighbor video retrieval on multiple video datasets and backbones. We also demonstrate significant improvement in fine-grained action classification for visually similar classes. With the commonly used 3D ResNet-18 architecture with UCF101 pretraining, we achieve 82.4\% (+5.1\% increase over the previous best) top-1 accuracy on UCF101 and 52.9\% (+5.4\% increase) on HMDB51 action classification, and 56.2\% (+11.7\% increase) Top-1 Recall on UCF101 nearest neighbor video retrieval. Code released at github.com/DAVEISHAN/TCLR.
Human action recognition is an active research area in computer vision. Although great process has been made, previous methods mostly recognize actions based on depth data at only one scale, and thus they often neglect multi-scale features that provide additional information action recognition in practical application scenarios. In this paper, we present a novel framework focusing on multi-scale motion information to recognize human actions from depth video sequences. We propose a multi-scale feature map called Laplacian pyramid depth motion images(LP-DMI). We employ depth motion images (DMI) as the templates to generate the multi-scale static representation of actions. Then, we caculate LP-DMI to enhance multi-scale dynamic information of motions and reduces redundant static information in human bodies. We further extract the multi-granularity descriptor called LP-DMI-HOG to provide more discriminative features. Finally, we utilize extreme learning machine (ELM) for action classification. The proposed method yeilds the recognition accuracy of 93.41%, 85.12%, 91.94% on public MSRAction3D dataset, UTD-MHAD and DHA dataset. Through extensive experiments, we prove that our method outperforms state-of-the-art benchmarks.
Temporal modeling still remains challenging for action recognition in videos. To mitigate this issue, this paper presents a new video architecture, termed as Temporal Difference Network (TDN), with a focus on capturing multi-scale temporal information for efficient action recognition. The core of our TDN is to devise an efficient temporal module (TDM) by explicitly leveraging a temporal difference operator, and systematically assess its effect on short-term and long-term motion modeling. To fully capture temporal information over the entire video, our TDN is established with a two-level difference modeling paradigm. Specifically, for local motion modeling, temporal difference over consecutive frames is used to supply 2D CNNs with finer motion pattern, while for global motion modeling, temporal difference across segments is incorporated to capture long-range structure for motion feature excitation. TDN provides a simple and principled temporal modeling framework and could be instantiated with the existing CNNs at a small extra computational cost. Our TDN presents a new state of the art on the Something-Something V1 & V2 datasets and is on par with the best performance on the Kinetics-400 dataset. In addition, we conduct in-depth ablation studies and plot the visualization results of our TDN, hopefully providing insightful analysis on temporal difference modeling. We release the code at https://github.com/MCG-NJU/TDN.
3D Convolutional Neural Network (3D CNN) captures spatial and temporal information on 3D data such as video sequences. However, due to the convolution and pooling mechanism, the information loss seems unavoidable. To improve the visual explanations and classification in 3D CNN, we propose two approaches; i) aggregate layer-wise global to local (global-local) discrete gradients using trained 3DResNext network, and ii) implement attention gating network to improve the accuracy of the action recognition. The proposed approach intends to show the usefulness of every layer termed as global-local attention in 3D CNN via visual attribution, weakly-supervised action localization, and action recognition. Firstly, the 3DResNext is trained and applied for action classification using backpropagation concerning the maximum predicted class. The gradients and activations of every layer are then up-sampled. Later, aggregation is used to produce more nuanced attention, which points out the most critical part of the predicted class's input videos. We use contour thresholding of final attention for final localization. We evaluate spatial and temporal action localization in trimmed videos using fine-grained visual explanation via 3DCam. Experimental results show that the proposed approach produces informative visual explanations and discriminative attention. Furthermore, the action recognition via attention gating on each layer produces better classification results than the baseline model.
Conventionally, spatiotemporal modeling network and its complexity are the two most concentrated research topics in video action recognition. Existing state-of-the-art methods have achieved excellent accuracy regardless of the complexity meanwhile efficient spatiotemporal modeling solutions are slightly inferior in performance. In this paper, we attempt to acquire both efficiency and effectiveness simultaneously. First of all, besides traditionally treating H x W x T video frames as space-time signal (viewing from the Height-Width spatial plane), we propose to also model video from the other two Height-Time and Width-Time planes, to capture the dynamics of video thoroughly. Secondly, our model is designed based on 2D CNN backbones and model complexity is well kept in mind by design. Specifically, we introduce a novel multi-view fusion (MVF) module to exploit video dynamics using separable convolution for efficiency. It is a plug-and-play module and can be inserted into off-the-shelf 2D CNNs to form a simple yet effective model called MVFNet. Moreover, MVFNet can be thought of as a generalized video modeling framework and it can specialize to be existing methods such as C2D, SlowOnly, and TSM under different settings. Extensive experiments are conducted on popular benchmarks (i.e., Something-Something V1 & V2, Kinetics, UCF-101, and HMDB-51) to show its superiority. The proposed MVFNet can achieve state-of-the-art performance with 2D CNN's complexity.
The paper addresses the problem of recognition of actions in video with low inter-class variability such as Table Tennis strokes. Two stream, "twin" convolutional neural networks are used with 3D convolutions both on RGB data and optical flow. Actions are recognized by classification of temporal windows. We introduce 3D attention modules and examine their impact on classification efficiency. In the context of the study of sportsmen performances, a corpus of the particular actions of table tennis strokes is considered. The use of attention blocks in the network speeds up the training step and improves the classification scores up to 5% with our twin model. We visualize the impact on the obtained features and notice correlation between attention and player movements and position. Score comparison of state-of-the-art action classification method and proposed approach with attentional blocks is performed on the corpus. Proposed model with attention blocks outperforms previous model without them and our baseline.
Most state-of-the-art methods for action recognition rely only on 2D spatial features encoding appearance, motion or pose. However, 2D data lacks the depth information, which is crucial for recognizing fine-grained actions. In this paper, we propose a depth-aware volumetric descriptor that encodes pose and motion information in a unified representation for action classification in-the-wild. Our framework is robust to many challenges inherent to action recognition, e.g. variation in viewpoint, scene, clothing and body shape. The key component of our method is the Depth-Aware Pose Motion representation (DA-PoTion), a new video descriptor that encodes the 3D movement of semantic keypoints of the human body. Given a video, we produce human joint heatmaps for each frame using a state-of-the-art 3D human pose regressor and we give each of them a unique color code according to the relative time in the clip. Then, we aggregate such 3D time-encoded heatmaps for all human joints to obtain a fixed-size descriptor (DA-PoTion), which is suitable for classifying actions using a shallow 3D convolutional neural network (CNN). The DA-PoTion alone defines a new state-of-the-art on the Penn Action Dataset. Moreover, we leverage the intrinsic complementarity of our pose motion descriptor with appearance based approaches by combining it with Inflated 3D ConvNet (I3D) to define a new state-of-the-art on the JHMDB Dataset.
Due to the large memory footprint of untrimmed videos, current state-of-the-art video localization methods operate atop precomputed video clip features. These features are extracted from video encoders typically trained for trimmed action classification tasks, making such features not necessarily suitable for temporal localization. In this work, we propose a novel supervised pretraining paradigm for clip features that not only trains to classify activities but also considers background clips and global video information to improve temporal sensitivity. Extensive experiments show that using features trained with our novel pretraining strategy significantly improves the performance of recent state-of-the-art methods on three tasks: Temporal Action Localization, Action Proposal Generation, and Dense Video Captioning. We also show that our pretraining approach is effective across three encoder architectures and two pretraining datasets. We believe video feature encoding is an important building block for localization algorithms, and extracting temporally-sensitive features should be of paramount importance in building more accurate models. The code and pretrained models are available on our project website.
Many video analysis tasks require temporal localization thus detection of content changes. However, most existing models developed for these tasks are pre-trained on general video action classification tasks. This is because large scale annotation of temporal boundaries in untrimmed videos is expensive. Therefore no suitable datasets exist for temporal boundary-sensitive pre-training. In this paper for the first time, we investigate model pre-training for temporal localization by introducing a novel boundary-sensitive pretext (BSP) task. Instead of relying on costly manual annotations of temporal boundaries, we propose to synthesize temporal boundaries in existing video action classification datasets. With the synthesized boundaries, BSP can be simply conducted via classifying the boundary types. This enables the learning of video representations that are much more transferable to downstream temporal localization tasks. Extensive experiments show that the proposed BSP is superior and complementary to the existing action classification based pre-training counterpart, and achieves new state-of-the-art performance on several temporal localization tasks.
Scene flow is the task of estimating 3D motion vectors to individual points of a dynamic 3D scene. Motion vectors have shown to be beneficial for downstream tasks such as action classification and collision avoidance. However, data collected via LiDAR sensors and stereo cameras are computation and labor intensive to precisely annotate for scene flow. We address this annotation bottleneck on two ends. We propose a 3D scene flow benchmark and a novel self-supervised setup for training flow models. The benchmark consists of datasets designed to study individual aspects of flow estimation in progressive order of complexity, from a single object in motion to real-world scenes. Furthermore, we introduce Adversarial Metric Learning for self-supervised flow estimation. The flow model is fed with sequences of point clouds to perform flow estimation. A second model learns a latent metric to distinguish between the points translated by the flow estimations and the target point cloud. This latent metric is learned via a Multi-Scale Triplet loss, which uses intermediary feature vectors for the loss calculation. We use our proposed benchmark to draw insights about the performance of the baselines and of different models when trained using our setup. We find that our setup is able to keep motion coherence and preserve local geometries, which many self-supervised baselines fail to grasp. Dealing with occlusions, on the other hand, is still an open challenge.
In this paper, we propose a study on multi-modal (audio and video) action spotting and classification in soccer videos. Action spotting and classification are the tasks that consist in finding the temporal anchors of events in a video and determine which event they are. This is an important application of general activity understanding. Here, we propose an experimental study on combining audio and video information at different stages of deep neural network architectures. We used the SoccerNet benchmark dataset, which contains annotated events for 500 soccer game videos from the Big Five European leagues. Through this work, we evaluated several ways to integrate audio stream into video-only-based architectures. We observed an average absolute improvement of the mean Average Precision (mAP) metric of $7.43\%$ for the action classification task and of $4.19\%$ for the action spotting task.
The construction of models for video action classification progresses rapidly. However, the performance of those models can still be easily improved by ensembling with the same models trained on different modalities (e.g. Optical flow). Unfortunately, it is computationally expensive to use several modalities during inference. Recent works examine the ways to integrate advantages of multi-modality into a single RGB-model. Yet, there is still a room for improvement. In this paper, we explore the various methods to embed the ensemble power into a single model. We show that proper initialization, as well as mutual modality learning, enhances single-modality models. As a result, we achieve state-of-the-art results in the Something-Something-v2 benchmark.
Recent progress on action recognition has mainly focused on RGB and optical flow features. In this paper, we approach the problem of joint-based action recognition. Unlike other modalities, constellation of joints and their motion generate models with succinct human motion information for activity recognition. We present a new model for joint-based action recognition, which first extracts motion features from each joint separately through a shared motion encoder before performing collective reasoning. Our joint selector module re-weights the joint information to select the most discriminative joints for the task. We also propose a novel joint-contrastive loss that pulls together groups of joint features which convey the same action. We strengthen the joint-based representations by using a geometry-aware data augmentation technique which jitters pose heatmaps while retaining the dynamics of the action. We show large improvements over the current state-of-the-art joint-based approaches on JHMDB, HMDB, Charades, AVA action recognition datasets. A late fusion with RGB and Flow-based approaches yields additional improvements. Our model also outperforms the existing baseline on Mimetics, a dataset with out-of-context actions.
In this paper we show that learning video feature spaces in which temporal cycles are maximally predictable benefits action classification. In particular, we propose a novel learning approach termed Cycle Encoding Prediction (CEP) that is able to effectively represent high-level spatio-temporal structure of unlabelled video content. CEP builds a latent space wherein the concept of closed forward-backward as well as backward-forward temporal loops is approximately preserved. As a self-supervision signal, CEP leverages the bi-directional temporal coherence of the video stream and applies loss functions that encourage both temporal cycle closure as well as contrastive feature separation. Architecturally, the underpinning network structure utilises a single feature encoder for all video snippets, adding two predictive modules that learn temporal forward and backward transitions. We apply our framework for pretext training of networks for action recognition tasks. We report significantly improved results for the standard datasets UCF101 and HMDB51. Detailed ablation studies support the effectiveness of the proposed components. We publish source code for the CEP components in full with this paper.
Convolutional Neural Networks with 3D kernels (3D-CNNs) currently achieve state-of-the-art results in video recognition tasks due to their supremacy in extracting spatiotemporal features within video frames. There have been many successful 3D-CNN architectures surpassing the state-of-the-art results successively. However, nearly all of them are designed to operate offline creating several serious handicaps during online operation. Firstly, conventional 3D-CNNs are not dynamic since their output features represent the complete input clip instead of the most recent frame in the clip. Secondly, they are not temporal resolution-preserving due to their inherent temporal downsampling. Lastly, 3D-CNNs are constrained to be used with fixed temporal input size limiting their flexibility. In order to address these drawbacks, we propose dissected 3D-CNNs, where the intermediate volumes of the network are dissected and propagated over depth (time) dimension for future calculations, substantially reducing the number of computations at online operation. For action classification, the dissected version of ResNet models performs 77-90% fewer computations at online operation while achieving ~5% better classification accuracy on the Kinetics-600 dataset than conventional 3D-ResNet models. Moreover, the advantages of dissected 3D-CNNs are demonstrated by deploying our approach onto several vision tasks, which consistently improved the performance.
We introduce Knowledge Fusion Transformers for video action classification. We present a self-attention based feature enhancer to fuse action knowledge in 3D inception based spatio-temporal context of the video clip intended to be classified. We show, how using only one stream networks and with little or, no pretraining can pave the way for a performance close to the current state-of-the-art. Additionally, we present how different self-attention architectures used at different levels of the network can be blended-in to enhance feature representation. Our architecture is trained and evaluated on UCF-101 and Charades dataset, where it is competitive with the state of the art. It also exceeds by a large gap from single stream networks with no to less pretraining.
In recent years, many works in the video action recognition literature have shown that two stream models (combining spatial and temporal input streams) are necessary for achieving state of the art performance. In this paper we show the benefits of including yet another stream based on human pose estimated from each frame -- specifically by rendering pose on input RGB frames. At first blush, this additional stream may seem redundant given that human pose is fully determined by RGB pixel values -- however we show (perhaps surprisingly) that this simple and flexible addition can provide complementary gains. Using this insight, we then propose a new model, which we dub PERF-Net (short for Pose Empowered RGB-Flow Net), which combines this new pose stream with the standard RGB and flow based input streams via distillation techniques and show that our model outperforms the state-of-the-art by a large margin in a number of human action recognition datasets while not requiring flow or pose to be explicitly computed at inference time. The proposed pose stream is also part of the winner solution of the ActivityNet Kinetics Challenge 2020.
We contribute HAA500, a manually annotated human-centric atomic action dataset for action recognition on 500 classes with over 591K labeled frames. To minimize ambiguities in action classification, HAA500 consists of highly diversified classes of fine-grained atomic actions, where only consistent actions fall under the same label, e.g., "Baseball Pitching" vs "Free Throw in Basketball". Thus HAA500 is different from existing atomic action datasets, where coarse-grained atomic actions were labeled with coarse action-verbs such as "Throw". HAA500 has been carefully curated to capture the precise movement of human figures with little class-irrelevant motions or spatio-temporal label noises. The advantages of HAA500 are fourfold: 1) human-centric actions with a high average of 69.7% detectable joints for the relevant human poses; 2) high scalability since adding a new class can be done under 20-60 minutes; 3) curated videos capturing essential elements of an atomic action without irrelevant frames; 4) fine-grained atomic action classes. Our extensive experiments including cross-data validation using datasets collected in the wild demonstrate the clear benefits of human-centric and atomic characteristics of HAA500, which enable training even a baseline deep learning model to improve prediction by attending to atomic human poses. We detail the HAA500 dataset statistics and collection methodology and compare quantitatively with existing action recognition datasets.