Novel Categories Discovery (NCD) facilitates learning from a partially annotated label space and enables deep learning (DL) models to operate in an open-world setting by identifying and differentiating instances of novel classes based on the labeled data notions. One of the primary assumptions of NCD is that the novel label space is perfectly disjoint and can be equipartitioned, but it is rarely realized by most NCD approaches in practice. To better align with this assumption, we propose a novel single-stage joint optimization-based NCD method, Negative learning, Entropy, and Variance regularization NCD (NEV-NCD). We demonstrate the efficacy of NEV-NCD in previously unexplored NCD applications of video action recognition (VAR) with the public UCF101 dataset and a curated in-house partial action-space annotated multi-view video dataset. We perform a thorough ablation study by varying the composition of final joint loss and associated hyper-parameters. During our experiments with UCF101 and multi-view action dataset, NEV-NCD achieves ~ 83% classification accuracy in test instances of labeled data. NEV-NCD achieves ~ 70% clustering accuracy over unlabeled data outperforming both naive baselines (by ~ 40%) and state-of-the-art pseudo-labeling-based approaches (by ~ 3.5%) over both datasets. Further, we propose to incorporate optional view-invariant feature learning with the multiview dataset to identify novel categories from novel viewpoints. Our additional view-invariance constraint improves the discriminative accuracy for both known and unknown categories by ~ 10% for novel viewpoints.
The task of weakly supervised temporal action localization targets at generating temporal boundaries for actions of interest, meanwhile the action category should also be classified. Pseudo-label-based methods, which serve as an effective solution, have been widely studied recently. However, existing methods generate pseudo labels during training and make predictions during testing under different pipelines or settings, resulting in a gap between training and testing. In this paper, we propose to generate high-quality pseudo labels from the predicted action boundaries. Nevertheless, we note that existing post-processing, like NMS, would lead to information loss, which is insufficient to generate high-quality action boundaries. More importantly, transforming action boundaries into pseudo labels is quite challenging, since the predicted action instances are generally overlapped and have different confidence scores. Besides, the generated pseudo-labels can be fluctuating and inaccurate at the early stage of training. It might repeatedly strengthen the false predictions if there is no mechanism to conduct self-correction. To tackle these issues, we come up with an effective pipeline for learning better pseudo labels. Firstly, we propose a Gaussian weighted fusion module to preserve information of action instances and obtain high-quality action boundaries. Second, we formulate the pseudo-label generation as an optimization problem under the constraints in terms of the confidence scores of action instances. Finally, we introduce the idea of $\Delta$ pseudo labels, which enables the model with the ability of self-correction. Our method achieves superior performance to existing methods on two benchmarks, THUMOS14 and ActivityNet1.3, achieving gains of 1.9\% on THUMOS14 and 3.7\% on ActivityNet1.3 in terms of average mAP.
We present a new algorithm for selection of informative frames in video action recognition. Our approach is designed for aerial videos captured using a moving camera where human actors occupy a small spatial resolution of video frames. Our algorithm utilizes the motion bias within aerial videos, which enables the selection of motion-salient frames. We introduce the concept of patch mutual information (PMI) score to quantify the motion bias between adjacent frames, by measuring the similarity of patches. We use this score to assess the amount of discriminative motion information contained in one frame relative to another. We present an adaptive frame selection strategy using shifted leaky ReLu and cumulative distribution function, which ensures that the sampled frames comprehensively cover all the essential segments with high motion salience. Our approach can be integrated with any action recognition model to enhance its accuracy. In practice, our method achieves a relative improvement of 2.2 - 13.8% in top-1 accuracy on UAV-Human, 6.8% on NEC Drone, and 9.0% on Diving48 datasets.
Research has shown the complementarity of camera- and inertial-based data for modeling human activities, yet datasets with both egocentric video and inertial-based sensor data remain scarce. In this paper, we introduce WEAR, an outdoor sports dataset for both vision- and inertial-based human activity recognition (HAR). Data from 22 participants performing a total of 18 different workout activities was collected with synchronized inertial (acceleration) and camera (egocentric video) data recorded at 11 different outside locations. WEAR provides a challenging prediction scenario in changing outdoor environments using a sensor placement, in line with recent trends in real-world applications. Benchmark results show that through our sensor placement, each modality interestingly offers complementary strengths and weaknesses in their prediction performance. Further, in light of the recent success of single-stage Temporal Action Localization (TAL) models, we demonstrate their versatility of not only being trained using visual data, but also using raw inertial data and being capable to fuse both modalities by means of simple concatenation. The dataset and code to reproduce experiments is publicly available via: mariusbock.github.io/wear/.
Monitoring the movement and actions of humans in video in real-time is an important task. We present a deep learning based algorithm for human action recognition for both RGB and thermal cameras. It is able to detect and track humans and recognize four basic actions (standing, walking, running, lying) in real-time on a notebook with a NVIDIA GPU. For this, it combines state of the art components for object detection (Scaled YoloV4), optical flow (RAFT) and pose estimation (EvoSkeleton). Qualitative experiments on a set of tunnel videos show that the proposed algorithm works robustly for both RGB and thermal video.
In this work we study the benefits of using tracking and 3D poses for action recognition. To achieve this, we take the Lagrangian view on analysing actions over a trajectory of human motion rather than at a fixed point in space. Taking this stand allows us to use the tracklets of people to predict their actions. In this spirit, first we show the benefits of using 3D pose to infer actions, and study person-person interactions. Subsequently, we propose a Lagrangian Action Recognition model by fusing 3D pose and contextualized appearance over tracklets. To this end, our method achieves state-of-the-art performance on the AVA v2.2 dataset on both pose only settings and on standard benchmark settings. When reasoning about the action using only pose cues, our pose model achieves +10.0 mAP gain over the corresponding state-of-the-art while our fused model has a gain of +2.8 mAP over the best state-of-the-art model. Code and results are available at: https://brjathu.github.io/LART
We study the problem of human action recognition using motion capture (MoCap) sequences. Unlike existing techniques that take multiple manual steps to derive standardized skeleton representations as model input, we propose a novel Spatial-Temporal Mesh Transformer (STMT) to directly model the mesh sequences. The model uses a hierarchical transformer with intra-frame off-set attention and inter-frame self-attention. The attention mechanism allows the model to freely attend between any two vertex patches to learn non-local relationships in the spatial-temporal domain. Masked vertex modeling and future frame prediction are used as two self-supervised tasks to fully activate the bi-directional and auto-regressive attention in our hierarchical transformer. The proposed method achieves state-of-the-art performance compared to skeleton-based and point-cloud-based models on common MoCap benchmarks. Code is available at https://github.com/zgzxy001/STMT.
Weakly-supervised temporal action localization aims to localize action instances in untrimmed videos with only video-level supervision. We witness that different actions record common phases, e.g., the run-up in the HighJump and LongJump. These different actions are defined as conjoint actions, whose rest parts are definite phases, e.g., leaping over the bar in a HighJump. Compared with the common phases, the definite phases are more easily localized in existing researches. Most of them formulate this task as a Multiple Instance Learning paradigm, in which the common phases are tended to be confused with the background, and affect the localization completeness of the conjoint actions. To tackle this challenge, we propose a Joint of Common and Definite phases Network (JCDNet) by improving feature discriminability of the conjoint actions. Specifically, we design a Class-Aware Discriminative module to enhance the contribution of the common phases in classification by the guidance of the coarse definite-phase features. Besides, we introduce a temporal attention module to learn robust action-ness scores via modeling temporal dependencies, distinguishing the common phases from the background. Extensive experiments on three datasets (THUMOS14, ActivityNetv1.2, and a conjoint-action subset) demonstrate that JCDNet achieves competitive performance against the state-of-the-art methods. Keywords: weakly-supervised learning, temporal action localization, conjoint action
For Human Action Recognition tasks (HAR), 3D Convolutional Neural Networks have proven to be highly effective, achieving state-of-the-art results. This study introduces a novel streaming architecture based toolflow for mapping such models onto FPGAs considering the model's inherent characteristics and the features of the targeted FPGA device. The HARFLOW3D toolflow takes as input a 3D CNN in ONNX format and a description of the FPGA characteristics, generating a design that minimizes the latency of the computation. The toolflow is comprised of a number of parts, including i) a 3D CNN parser, ii) a performance and resource model, iii) a scheduling algorithm for executing 3D models on the generated hardware, iv) a resource-aware optimization engine tailored for 3D models, v) an automated mapping to synthesizable code for FPGAs. The ability of the toolflow to support a broad range of models and devices is shown through a number of experiments on various 3D CNN and FPGA system pairs. Furthermore, the toolflow has produced high-performing results for 3D CNN models that have not been mapped to FPGAs before, demonstrating the potential of FPGA-based systems in this space. Overall, HARFLOW3D has demonstrated its ability to deliver competitive latency compared to a range of state-of-the-art hand-tuned approaches being able to achieve up to 5$\times$ better performance compared to some of the existing works.
Existing video-language pre-training methods primarily focus on instance-level alignment between video clips and captions via global contrastive learning but neglect rich fine-grained local information in both videos and text, which is of importance to downstream tasks requiring temporal localization and semantic reasoning. A powerful model is expected to be capable of capturing region-object correspondences and recognizing scene changes in a video clip, reflecting spatial and temporal granularity, respectively. To strengthen model's understanding into such fine-grained details, we propose a simple yet effective video-language modeling framework, S-ViLM, by exploiting the intrinsic structures of these two modalities. It includes two novel designs, inter-clip spatial grounding and intra-clip temporal grouping, to promote learning region-object alignment and temporal-aware features, simultaneously. Comprehensive evaluations demonstrate that S-ViLM performs favorably against existing approaches in learning more expressive representations. Specifically, S-ViLM surpasses the state-of-the-art methods substantially on four representative downstream tasks, covering text-video retrieval, video question answering, video action recognition, and temporal action localization.
Scale is the primary factor for building a powerful foundation model that could well generalize to a variety of downstream tasks. However, it is still challenging to train video foundation models with billions of parameters. This paper shows that video masked autoencoder (VideoMAE) is a scalable and general self-supervised pre-trainer for building video foundation models. We scale the VideoMAE in both model and data with a core design. Specifically, we present a dual masking strategy for efficient pre-training, with an encoder operating on a subset of video tokens and a decoder processing another subset of video tokens. Although VideoMAE is very efficient due to high masking ratio in encoder, masking decoder can still further reduce the overall computational cost. This enables the efficient pre-training of billion-level models in video. We also use a progressive training paradigm that involves an initial pre-training on a diverse multi-sourced unlabeled dataset, followed by a post-pre-training on a mixed labeled dataset. Finally, we successfully train a video ViT model with a billion parameters, which achieves a new state-of-the-art performance on the datasets of Kinetics (90.0% on K400 and 89.9% on K600) and Something-Something (68.7% on V1 and 77.0% on V2). In addition, we extensively verify the pre-trained video ViT models on a variety of downstream tasks, demonstrating its effectiveness as a general video representation learner. The code and model is available at \url{https://github.com/OpenGVLab/VideoMAEv2}.
Recently, Transformer-based methods have been utilized to improve the performance of human action recognition. However, most of these studies assume that multi-view data is complete, which may not always be the case in real-world scenarios. Therefore, this paper presents a novel Multi-view Knowledge Distillation Transformer (MKDT) framework that consists of a teacher network and a student network. This framework aims to handle incomplete human action problems in real-world applications. Specifically, the multi-view knowledge distillation transformer uses a hierarchical vision transformer with shifted windows to capture more spatial-temporal information. Experimental results demonstrate that our framework outperforms the CNN-based method on three public datasets.
This paper simultaneously addresses three limitations associated with conventional skeleton-based action recognition; skeleton detection and tracking errors, poor variety of the targeted actions, as well as person-wise and frame-wise action recognition. A point cloud deep-learning paradigm is introduced to the action recognition, and a unified framework along with a novel deep neural network architecture called Structured Keypoint Pooling is proposed. The proposed method sparsely aggregates keypoint features in a cascaded manner based on prior knowledge of the data structure (which is inherent in skeletons), such as the instances and frames to which each keypoint belongs, and achieves robustness against input errors. Its less constrained and tracking-free architecture enables time-series keypoints consisting of human skeletons and nonhuman object contours to be efficiently treated as an input 3D point cloud and extends the variety of the targeted action. Furthermore, we propose a Pooling-Switching Trick inspired by Structured Keypoint Pooling. This trick switches the pooling kernels between the training and inference phases to detect person-wise and frame-wise actions in a weakly supervised manner using only video-level action labels. This trick enables our training scheme to naturally introduce novel data augmentation, which mixes multiple point clouds extracted from different videos. In the experiments, we comprehensively verify the effectiveness of the proposed method against the limitations, and the method outperforms state-of-the-art skeleton-based action recognition and spatio-temporal action localization methods.
The goal of building a benchmark (suite of datasets) is to provide a unified protocol for fair evaluation and thus facilitate the evolution of a specific area. Nonetheless, we point out that existing protocols of action recognition could yield partial evaluations due to several limitations. To comprehensively probe the effectiveness of spatiotemporal representation learning, we introduce BEAR, a new BEnchmark on video Action Recognition. BEAR is a collection of 18 video datasets grouped into 5 categories (anomaly, gesture, daily, sports, and instructional), which covers a diverse set of real-world applications. With BEAR, we thoroughly evaluate 6 common spatiotemporal models pre-trained by both supervised and self-supervised learning. We also report transfer performance via standard finetuning, few-shot finetuning, and unsupervised domain adaptation. Our observation suggests that current state-of-the-art cannot solidly guarantee high performance on datasets close to real-world applications, and we hope BEAR can serve as a fair and challenging evaluation benchmark to gain insights on building next-generation spatiotemporal learners. Our dataset, code, and models are released at: https://github.com/AndongDeng/BEAR
Weakly-supervised temporal action localization aims to locate action regions and identify action categories in untrimmed videos simultaneously by taking only video-level labels as the supervision. Pseudo label generation is a promising strategy to solve the challenging problem, but the current methods ignore the natural temporal structure of the video that can provide rich information to assist such a generation process. In this paper, we propose a novel weakly-supervised temporal action localization method by inferring salient snippet-feature. First, we design a saliency inference module that exploits the variation relationship between temporal neighbor snippets to discover salient snippet-features, which can reflect the significant dynamic change in the video. Secondly, we introduce a boundary refinement module that enhances salient snippet-features through the information interaction unit. Then, a discrimination enhancement module is introduced to enhance the discriminative nature of snippet-features. Finally, we adopt the refined snippet-features to produce high-fidelity pseudo labels, which could be used to supervise the training of the action localization network. Extensive experiments on two publicly available datasets, i.e., THUMOS14 and ActivityNet v1.3, demonstrate our proposed method achieves significant improvements compared to the state-of-the-art methods.
In this paper, we consider the problem of temporal action localization under low-shot (zero-shot & few-shot) scenario, with the goal of detecting and classifying the action instances from arbitrary categories within some untrimmed videos, even not seen at training time. We adopt a Transformer-based two-stage action localization architecture with class-agnostic action proposal, followed by open-vocabulary classification. We make the following contributions. First, to compensate image-text foundation models with temporal motions, we improve category-agnostic action proposal by explicitly aligning embeddings of optical flows, RGB and texts, which has largely been ignored in existing low-shot methods. Second, to improve open-vocabulary action classification, we construct classifiers with strong discriminative power, i.e., avoid lexical ambiguities. To be specific, we propose to prompt the pre-trained CLIP text encoder either with detailed action descriptions (acquired from large-scale language models), or visually-conditioned instance-specific prompt vectors. Third, we conduct thorough experiments and ablation studies on THUMOS14 and ActivityNet1.3, demonstrating the superior performance of our proposed model, outperforming existing state-of-the-art approaches by one significant margin.
Human action recognition is a challenging problem, particularly when there is high variability in factors such as subject appearance, backgrounds and viewpoint. While deep neural networks (DNNs) have been shown to perform well on action recognition tasks, they typically require large amounts of high-quality labeled data to achieve robust performance across a variety of conditions. Synthetic data has shown promise as a way to avoid the substantial costs and potential ethical concerns associated with collecting and labeling enormous amounts of data in the real-world. However, synthetic data may differ from real data in important ways. This phenomenon, known as \textit{domain shift}, can limit the utility of synthetic data in robotics applications. To mitigate the effects of domain shift, substantial effort is being dedicated to the development of domain adaptation (DA) techniques. Yet, much remains to be understood about how best to develop these techniques. In this paper, we introduce a new dataset called Robot Control Gestures (RoCoG-v2). The dataset is composed of both real and synthetic videos from seven gesture classes, and is intended to support the study of synthetic-to-real domain shift for video-based action recognition. Our work expands upon existing datasets by focusing the action classes on gestures for human-robot teaming, as well as by enabling investigation of domain shift in both ground and aerial views. We present baseline results using state-of-the-art action recognition and domain adaptation algorithms and offer initial insight on tackling the synthetic-to-real and ground-to-air domain shifts.
Visual-Language Models (VLMs) have significantly advanced action video recognition. Supervised by the semantics of action labels, recent works adapt the visual branch of VLMs to learn video representations. Despite the effectiveness proved by these works, we believe that the potential of VLMs has yet to be fully harnessed. In light of this, we exploit the semantic units (SU) hiding behind the action labels and leverage their correlations with fine-grained items in frames for more accurate action recognition. SUs are entities extracted from the language descriptions of the entire action set, including body parts, objects, scenes, and motions. To further enhance the alignments between visual contents and the SUs, we introduce a multi-region module (MRA) to the visual branch of the VLM. The MRA allows the perception of region-aware visual features beyond the original global feature. Our method adaptively attends to and selects relevant SUs with visual features of frames. With a cross-modal decoder, the selected SUs serve to decode spatiotemporal video representations. In summary, the SUs as the medium can boost discriminative ability and transferability. Specifically, in fully-supervised learning, our method achieved 87.8% top-1 accuracy on Kinetics-400. In K=2 few-shot experiments, our method surpassed the previous state-of-the-art by +7.1% and +15.0% on HMDB-51 and UCF-101, respectively.
Temporal Action Localization (TAL) is a challenging task in video understanding that aims to identify and localize actions within a video sequence. Recent studies have emphasized the importance of applying long-term temporal context modeling (TCM) blocks to the extracted video clip features such as employing complex self-attention mechanisms. In this paper, we present the simplest method ever to address this task and argue that the extracted video clip features are already informative to achieve outstanding performance without sophisticated architectures. To this end, we introduce TemporalMaxer, which minimizes long-term temporal context modeling while maximizing information from the extracted video clip features with a basic, parameter-free, and local region operating max-pooling block. Picking out only the most critical information for adjacent and local clip embeddings, this block results in a more efficient TAL model. We demonstrate that TemporalMaxer outperforms other state-of-the-art methods that utilize long-term TCM such as self-attention on various TAL datasets while requiring significantly fewer parameters and computational resources. The code for our approach is publicly available at https://github.com/TuanTNG/TemporalMaxer
Temporal action localization (TAL) is a prevailing task due to its great application potential. Existing works in this field mainly suffer from two weaknesses: (1) They often neglect the multi-label case and only focus on temporal modeling. (2) They ignore the semantic information in class labels and only use the visual information. To solve these problems, we propose a novel Co-Occurrence Relation Module (CORM) that explicitly models the co-occurrence relationship between actions. Besides the visual information, it further utilizes the semantic embeddings of class labels to model the co-occurrence relationship. The CORM works in a plug-and-play manner and can be easily incorporated with the existing sequence models. By considering both visual and semantic co-occurrence, our method achieves high multi-label relationship modeling capacity. Meanwhile, existing datasets in TAL always focus on low-semantic atomic actions. Thus we construct a challenging multi-label dataset UCF-Crime-TAL that focuses on high-semantic actions by annotating the UCF-Crime dataset at frame level and considering the semantic overlap of different events. Extensive experiments on two commonly used TAL datasets, \textit{i.e.}, MultiTHUMOS and TSU, and our newly proposed UCF-Crime-TAL demenstrate the effectiveness of the proposed CORM, which achieves state-of-the-art performance on these datasets.