Video segmentation requires consistently segmenting and tracking objects over time. Due to the quadratic dependency on input size, directly applying self-attention to video segmentation with high-resolution input features poses significant challenges, often leading to insufficient GPU memory capacity. Consequently, modern video segmenters either extend an image segmenter without incorporating any temporal attention or resort to window space-time attention in a naive manner. In this work, we present Axial-VS, a general and simple framework that enhances video segmenters by tracking objects along axial trajectories. The framework tackles video segmentation through two sub-tasks: short-term within-clip segmentation and long-term cross-clip tracking. In the first step, Axial-VS augments an off-the-shelf clip-level video segmenter with the proposed axial-trajectory attention, sequentially tracking objects along the height- and width-trajectories within a clip, thereby enhancing temporal consistency by capturing motion trajectories. The axial decomposition significantly reduces the computational complexity for dense features, and outperforms the window space-time attention in segmentation quality. In the second step, we further employ axial-trajectory attention to the object queries in clip-level segmenters, which are learned to encode object information, thereby aiding object tracking across different clips and achieving consistent segmentation throughout the video. Without bells and whistles, Axial-VS showcases state-of-the-art results on video segmentation benchmarks, emphasizing its effectiveness in addressing the limitations of modern clip-level video segmenters. Code and models are available at https://github.com/TACJu/Axial-VS.
Deep learning-based models are at the forefront of most driver observation benchmarks due to their remarkable accuracies but are also associated with high computational costs. This is challenging, as resources are often limited in real-world driving scenarios. This paper introduces a lightweight framework for resource-efficient driver activity recognition. The framework enhances 3D MobileNet, a neural architecture optimized for speed in video classification, by incorporating knowledge distillation and model quantization to balance model accuracy and computational efficiency. Knowledge distillation helps maintain accuracy while reducing the model size by leveraging soft labels from a larger teacher model (I3D), instead of relying solely on original ground truth data. Model quantization significantly lowers memory and computation demands by using lower precision integers for model weights and activations. Extensive testing on a public dataset for in-vehicle monitoring during autonomous driving demonstrates that this new framework achieves a threefold reduction in model size and a 1.4-fold improvement in inference time, compared to an already optimized architecture. The code for this study is available at https://github.com/calvintanama/qd-driver-activity-reco.
Inspired by recent developments in attention models for image classification and natural language processing, we present various Attention based architectures in reinforcement learning (RL) domain, capable of performing well on OpenAI Gym Atari-2600 game suite. In spite of the recent success of Deep Reinforcement learning techniques in various fields like robotics, gaming and healthcare, they suffer from a major drawback that neural networks are difficult to interpret. We try to get around this problem with the help of Attention based models. In Attention based models, extracting and overlaying of attention map onto images allows for direct observation of information used by agent to select actions and easier interpretation of logic behind the chosen actions. Our models in addition to playing well on gym-Atari environments, also provide insights on how agent perceives its environment. In addition, motivated by recent developments in attention based video-classification models using Vision Transformer, we come up with an architecture based on Vision Transformer, for image-based RL domain too. Compared to previous works in Vision Transformer, our model is faster to train and requires fewer computational resources. 3
While most modern video understanding models operate on short-range clips, real-world videos are often several minutes long with semantically consistent segments of variable length. A common approach to process long videos is applying a short-form video model over uniformly sampled clips of fixed temporal length and aggregating the outputs. This approach neglects the underlying nature of long videos since fixed-length clips are often redundant or uninformative. In this paper, we aim to provide a generic and adaptive sampling approach for long-form videos in lieu of the de facto uniform sampling. Viewing videos as semantically consistent segments, we formulate a task-agnostic, unsupervised, and scalable approach based on Kernel Temporal Segmentation (KTS) for sampling and tokenizing long videos. We evaluate our method on long-form video understanding tasks such as video classification and temporal action localization, showing consistent gains over existing approaches and achieving state-of-the-art performance on long-form video modeling.
Despite an exciting new wave of multimodal machine learning models, current approaches still struggle to interpret the complex contextual relationships between the different modalities present in videos. Going beyond existing methods that emphasize simple activities or objects, we propose a new model-agnostic approach for generating detailed textual descriptions that captures multimodal video information. Our method leverages the extensive knowledge learnt by large language models, such as GPT-3.5 or Llama2, to reason about textual descriptions of the visual and aural modalities, obtained from BLIP-2, Whisper and ImageBind. Without needing additional finetuning of video-text models or datasets, we demonstrate that available LLMs have the ability to use these multimodal textual descriptions as proxies for ``sight'' or ``hearing'' and perform zero-shot multimodal classification of videos in-context. Our evaluations on popular action recognition benchmarks, such as UCF-101 or Kinetics, show these context-rich descriptions can be successfully used in video understanding tasks. This method points towards a promising new research direction in multimodal classification, demonstrating how an interplay between textual, visual and auditory machine learning models can enable more holistic video understanding.
Learning high-quality video representation has shown significant applications in computer vision and remains challenging. Previous work based on mask autoencoders such as ImageMAE and VideoMAE has proven the effectiveness of learning representations in images and videos through reconstruction strategy in the visual modality. However, these models exhibit inherent limitations, particularly in scenarios where extracting features solely from the visual modality proves challenging, such as when dealing with low-resolution and blurry original videos. Based on this, we propose AV-MaskEnhancer for learning high-quality video representation by combining visual and audio information. Our approach addresses the challenge by demonstrating the complementary nature of audio and video features in cross-modality content. Moreover, our result of the video classification task on the UCF101 dataset outperforms the existing work and reaches the state-of-the-art, with a top-1 accuracy of 98.8% and a top-5 accuracy of 99.9%.
Optimizing video inference efficiency has become increasingly important with the growing demand for video analysis in various fields. Some existing methods achieve high efficiency by explicit discard of spatial or temporal information, which poses challenges in fast-changing and fine-grained scenarios. To address these issues, we propose an efficient video representation network with Differentiable Resolution Compression and Alignment mechanism, which compresses non-essential information in the early stage of the network to reduce computational costs while maintaining consistent temporal correlations. Specifically, we leverage a Differentiable Context-aware Compression Module to encode the saliency and non-saliency frame features, refining and updating the features into a high-low resolution video sequence. To process the new sequence, we introduce a new Resolution-Align Transformer Layer to capture global temporal correlations among frame features with different resolutions, while reducing spatial computation costs quadratically by utilizing fewer spatial tokens in low-resolution non-saliency frames. The entire network can be end-to-end optimized via the integration of the differentiable compression module. Experimental results show that our method achieves the best trade-off between efficiency and performance on near-duplicate video retrieval and competitive results on dynamic video classification compared to state-of-the-art methods. Code:https://github.com/dun-research/DRCA
Training deep learning models for video classification from audio-visual data commonly requires immense amounts of labeled training data collected via a costly process. A challenging and underexplored, yet much cheaper, setup is few-shot learning from video data. In particular, the inherently multi-modal nature of video data with sound and visual information has not been leveraged extensively for the few-shot video classification task. Therefore, we introduce a unified audio-visual few-shot video classification benchmark on three datasets, i.e. the VGGSound-FSL, UCF-FSL, ActivityNet-FSL datasets, where we adapt and compare ten methods. In addition, we propose AV-DIFF, a text-to-feature diffusion framework, which first fuses the temporal and audio-visual features via cross-modal attention and then generates multi-modal features for the novel classes. We show that AV-DIFF obtains state-of-the-art performance on our proposed benchmark for audio-visual (generalised) few-shot learning. Our benchmark paves the way for effective audio-visual classification when only limited labeled data is available. Code and data are available at https://github.com/ExplainableML/AVDIFF-GFSL.
Misinformation on YouTube is a significant concern, necessitating robust detection strategies. In this paper, we introduce a novel methodology for video classification, focusing on the veracity of the content. We convert the conventional video classification task into a text classification task by leveraging the textual content derived from the video transcripts. We employ advanced machine learning techniques like transfer learning to solve the classification challenge. Our approach incorporates two forms of transfer learning: (a) fine-tuning base transformer models such as BERT, RoBERTa, and ELECTRA, and (b) few-shot learning using sentence-transformers MPNet and RoBERTa-large. We apply the trained models to three datasets: (a) YouTube Vaccine-misinformation related videos, (b) YouTube Pseudoscience videos, and (c) Fake-News dataset (a collection of articles). Including the Fake-News dataset extended the evaluation of our approach beyond YouTube videos. Using these datasets, we evaluated the models distinguishing valid information from misinformation. The fine-tuned models yielded Matthews Correlation Coefficient>0.81, accuracy>0.90, and F1 score>0.90 in two of three datasets. Interestingly, the few-shot models outperformed the fine-tuned ones by 20% in both Accuracy and F1 score for the YouTube Pseudoscience dataset, highlighting the potential utility of this approach -- especially in the context of limited training data.
Early diagnosis of renal cancer can greatly improve the survival rate of patients. Contrast-enhanced ultrasound (CEUS) is a cost-effective and non-invasive imaging technique and has become more and more frequently used for renal tumor diagnosis. However, the classification of benign and malignant renal tumors can still be very challenging due to the highly heterogeneous appearance of cancer and imaging artifacts. Our aim is to detect and classify renal tumors by integrating B-mode and CEUS-mode ultrasound videos. To this end, we propose a novel multi-modal ultrasound video fusion network that can effectively perform multi-modal feature fusion and video classification for renal tumor diagnosis. The attention-based multi-modal fusion module uses cross-attention and self-attention to extract modality-invariant features and modality-specific features in parallel. In addition, we design an object-level temporal aggregation (OTA) module that can automatically filter low-quality features and efficiently integrate temporal information from multiple frames to improve the accuracy of tumor diagnosis. Experimental results on a multicenter dataset show that the proposed framework outperforms the single-modal models and the competing methods. Furthermore, our OTA module achieves higher classification accuracy than the frame-level predictions. Our code is available at \url{https://github.com/JeunyuLi/MUAF}.
The ubiquitous and demonstrably suboptimal choice of resizing images to a fixed resolution before processing them with computer vision models has not yet been successfully challenged. However, models such as the Vision Transformer (ViT) offer flexible sequence-based modeling, and hence varying input sequence lengths. We take advantage of this with NaViT (Native Resolution ViT) which uses sequence packing during training to process inputs of arbitrary resolutions and aspect ratios. Alongside flexible model usage, we demonstrate improved training efficiency for large-scale supervised and contrastive image-text pretraining. NaViT can be efficiently transferred to standard tasks such as image and video classification, object detection, and semantic segmentation and leads to improved results on robustness and fairness benchmarks. At inference time, the input resolution flexibility can be used to smoothly navigate the test-time cost-performance trade-off. We believe that NaViT marks a departure from the standard, CNN-designed, input and modelling pipeline used by most computer vision models, and represents a promising direction for ViTs.
Deep learning algorithms have pushed the boundaries of computer vision research and have depicted commendable performance in a variety of applications. However, training a robust deep neural network necessitates a large amount of labeled training data, acquiring which involves significant time and human effort. This problem is even more serious for an application like video classification, where a human annotator has to watch an entire video end-to-end to furnish a label. Active learning algorithms automatically identify the most informative samples from large amounts of unlabeled data; this tremendously reduces the human annotation effort in inducing a machine learning model, as only the few samples that are identified by the algorithm, need to be labeled manually. In this paper, we propose a novel active learning framework for video classification, with the goal of further reducing the labeling onus on the human annotators. Our framework identifies a batch of exemplar videos, together with a set of informative frames for each video; the human annotator needs to merely review the frames and provide a label for each video. This involves much less manual work than watching the complete video to come up with a label. We formulate a criterion based on uncertainty and diversity to identify the informative videos and exploit representative sampling techniques to extract a set of exemplar frames from each video. To the best of our knowledge, this is the first research effort to develop an active learning framework for video classification, where the annotators need to inspect only a few frames to produce a label, rather than watching the end-to-end video.
In the context of label-efficient learning on video data, the distillation method and the structural design of the teacher-student architecture have a significant impact on knowledge distillation. However, the relationship between these factors has been overlooked in previous research. To address this gap, we propose a new weakly supervised learning framework for knowledge distillation in video classification that is designed to improve the efficiency and accuracy of the student model. Our approach leverages the concept of substage-based learning to distill knowledge based on the combination of student substages and the correlation of corresponding substages. We also employ the progressive cascade training method to address the accuracy loss caused by the large capacity gap between the teacher and the student. Additionally, we propose a pseudo-label optimization strategy to improve the initial data label. To optimize the loss functions of different distillation substages during the training process, we introduce a new loss method based on feature distribution. We conduct extensive experiments on both real and simulated data sets, demonstrating that our proposed approach outperforms existing distillation methods in terms of knowledge distillation for video classification tasks. Our proposed substage-based distillation approach has the potential to inform future research on label-efficient learning for video data.
Multimodal learning assumes all modality combinations of interest are available during training to learn cross-modal correspondences. In this paper, we challenge this modality-complete assumption for multimodal learning and instead strive for generalization to unseen modality combinations during inference. We pose the problem of unseen modality interaction and introduce a first solution. It exploits a module that projects the multidimensional features of different modalities into a common space with rich information preserved. This allows the information to be accumulated with a simple summation operation across available modalities. To reduce overfitting to less discriminative modality combinations during training, we further improve the model learning with pseudo-supervision indicating the reliability of a modality's prediction. We demonstrate that our approach is effective for diverse tasks and modalities by evaluating it for multimodal video classification, robot state regression, and multimedia retrieval. Project website: https://xiaobai1217.github.io/Unseen-Modality-Interaction/.
Breast ultrasound videos contain richer information than ultrasound images, therefore it is more meaningful to develop video models for this diagnosis task. However, the collection of ultrasound video datasets is much harder. In this paper, we explore the feasibility of enhancing the performance of ultrasound video classification using the static image dataset. To this end, we propose KGA-Net and coherence loss. The KGA-Net adopts both video clips and static images to train the network. The coherence loss uses the feature centers generated by the static images to guide the frame attention in the video model. Our KGA-Net boosts the performance on the public BUSV dataset by a large margin. The visualization results of frame attention prove the explainability of our method. The codes and model weights of our method will be made publicly available.
Localization of the narrowest position of the vessel and corresponding vessel and remnant vessel delineation in carotid ultrasound (US) are essential for carotid stenosis grading (CSG) in clinical practice. However, the pipeline is time-consuming and tough due to the ambiguous boundaries of plaque and temporal variation. To automatize this procedure, a large number of manual delineations are usually required, which is not only laborious but also not reliable given the annotation difficulty. In this study, we present the first video classification framework for automatic CSG. Our contribution is three-fold. First, to avoid the requirement of laborious and unreliable annotation, we propose a novel and effective video classification network for weakly-supervised CSG. Second, to ease the model training, we adopt an inflation strategy for the network, where pre-trained 2D convolution weights can be adapted into the 3D counterpart in our network for an effective warm start. Third, to enhance the feature discrimination of the video, we propose a novel attention-guided multi-dimension fusion (AMDF) transformer encoder to model and integrate global dependencies within and across spatial and temporal dimensions, where two lightweight cross-dimensional attention mechanisms are designed. Our approach is extensively validated on a large clinically collected carotid US video dataset, demonstrating state-of-the-art performance compared with strong competitors.
Today ship hull inspection including the examination of the external coating, detection of defects, and other types of external degradation such as corrosion and marine growth is conducted underwater by means of Remotely Operated Vehicles (ROVs). The inspection process consists of a manual video analysis which is a time-consuming and labor-intensive process. To address this, we propose an automatic video analysis system using deep learning and computer vision to improve upon existing methods that only consider spatial information on individual frames in underwater ship hull video inspection. By exploring the benefits of adding temporal information and analyzing frame-based classifiers, we propose a multi-label video classification model that exploits the self-attention mechanism of transformers to capture spatiotemporal attention in consecutive video frames. Our proposed method has demonstrated promising results and can serve as a benchmark for future research and development in underwater video inspection applications.
Online video platforms receive hundreds of hours of uploads every minute, making manual content moderation impossible. Unfortunately, the most vulnerable consumers of malicious video content are children from ages 1-5 whose attention is easily captured by bursts of color and sound. Scammers attempting to monetize their content may craft malicious children's videos that are superficially similar to educational videos, but include scary and disgusting characters, violent motions, loud music, and disturbing noises. Prominent video hosting platforms like YouTube have taken measures to mitigate malicious content on their platform, but these videos often go undetected by current content moderation tools that are focused on removing pornographic or copyrighted content. This paper introduces our toolkit Malicious or Benign for promoting research on automated content moderation of children's videos. We present 1) a customizable annotation tool for videos, 2) a new dataset with difficult to detect test cases of malicious content and 3) a benchmark suite of state-of-the-art video classification models.
We present Integrated Multimodal Perception (IMP), a simple and scalable multimodal multi-task training and modeling approach. IMP integrates multimodal inputs including image, video, text, and audio into a single Transformer encoder with minimal modality-specific components. IMP makes use of a novel design that combines Alternating Gradient Descent (AGD) and Mixture-of-Experts (MoE) for efficient model and task scaling. We conduct extensive empirical studies and reveal the following key insights: 1) Performing gradient descent updates by alternating on diverse modalities, loss functions, and tasks, with varying input resolutions, efficiently improves the model. 2) Sparsification with MoE on a single modality-agnostic encoder substantially improves the performance, outperforming dense models that use modality-specific encoders or additional fusion layers and greatly mitigates the conflicts between modalities. IMP achieves competitive performance on a wide range of downstream tasks including video classification, image classification, image-text, and video-text retrieval. Most notably, we train a sparse IMP-MoE-L variant focusing on video tasks that achieves new state-of-the-art in zero-shot video classification: 77.0% on Kinetics-400, 76.8% on Kinetics-600, and 68.3% on Kinetics-700, improving the previous state-of-the-art by +5%, +6.7%, and +5.8%, respectively, while using only 15% of their total training computational cost.
Hate speech has become one of the most significant issues in modern society, having implications in both the online and the offline world. Due to this, hate speech research has recently gained a lot of traction. However, most of the work has primarily focused on text media with relatively little work on images and even lesser on videos. Thus, early stage automated video moderation techniques are needed to handle the videos that are being uploaded to keep the platform safe and healthy. With a view to detect and remove hateful content from the video sharing platforms, our work focuses on hate video detection using multi-modalities. To this end, we curate ~43 hours of videos from BitChute and manually annotate them as hate or non-hate, along with the frame spans which could explain the labelling decision. To collect the relevant videos we harnessed search keywords from hate lexicons. We observe various cues in images and audio of hateful videos. Further, we build deep learning multi-modal models to classify the hate videos and observe that using all the modalities of the videos improves the overall hate speech detection performance (accuracy=0.798, macro F1-score=0.790) by ~5.7% compared to the best uni-modal model in terms of macro F1 score. In summary, our work takes the first step toward understanding and modeling hateful videos on video hosting platforms such as BitChute.