Video anomaly detection (VAD) aims at localizing unexpected actions or activities in a video sequence. Existing mainstream VAD techniques are based on either the one-class formulation, which assumes all training data are normal, or weakly-supervised, which requires only video-level normal/anomaly labels. To establish a unified approach to solving the two VAD settings, we introduce a self-supervised sparse representation (S3R) framework that models the concept of anomaly at feature level by exploring the synergy between dictionary-based representation and self-supervised learning. With the learned dictionary, S3R facilitates two coupled modules, en-Normal and de-Normal, to reconstruct snippet-level features and filter out normal-event features. The self-supervised techniques also enable generating samples of pseudo normal/anomaly to train the anomaly detector. We demonstrate with extensive experiments that S3R achieves new state-of-the-art performances on popular benchmark datasets for both one-class and weakly-supervised VAD tasks. Our code is publicly available at https://github.com/louisYen/S3R.
论文检索
输入标题、作者或关键词，从 8,216 篇学术成果中精准定位
Trustworthy machine learning is of primary importance to the practical deployment of deep learning models. While state-of-the-art models achieve astonishingly good performance in terms of accuracy, recent literature reveals that their predictive confidence scores unfortunately cannot be trusted: e.g., they are often overconfident when wrong predictions are made, or so even for obvious outliers. In this paper, we introduce a new approach of \emph{self-supervised probing}, which enables us to check and mitigate the overconfidence issue for a trained model, thereby improving its trustworthiness. We provide a simple yet effective framework, which can be flexibly applied to existing trustworthiness-related methods in a plug-and-play manner. Extensive experiments on three trustworthiness-related tasks (misclassification detection, calibration and out-of-distribution detection) across various benchmarks verify the effectiveness of our proposed probing framework.
We present a new on-road driving dataset, called “Look Both Ways”, which contains synchronized video of both driver faces and the forward road scene, along with ground truth gaze data registered from eye tracking glasses worn by the drivers. Our dataset supports the study of methods for non-intrusively estimating a driver’s focus of attention while driving - an important application area in road safety. A key challenge is that this task requires accurate gaze estimation, but supervised appearance-based gaze estimation methods often do not transfer well to real driving datasets, and in-domain ground truth to supervise them is difficult to gather. We therefore propose a method for self-supervision of driver gaze, by taking advantage of the geometric consistency between the driver’s gaze direction and the saliency of the scene as observed by the driver. We formulate a 3D geometric learning framework to enforce this consistency, allowing the gaze model to supervise the scene saliency model, and vice versa. We implement a prototype of our method and test it with our dataset, to show that compared to a supervised approach it can yield better gaze estimation and scene saliency estimation with no additional labels.
Towards Unbiased Label Distribution Learning for Facial Pose Estimation Using Anisotropic Spherical Gaussian
PDF ↗Facial pose estimation refers to the task of predicting face orientation from a single RGB image. It is an important research topic with a wide range of applications in computer vision. Label distribution learning (LDL) based methods have been recently proposed for facial pose estimation, which achieve promising results. However, there are two major issues in existing LDL methods. First, the expectations of label distributions are biased, leading to a biased pose estimation. Second, fixed distribution parameters are applied for all learning samples, severely limiting the model capability. In this paper, we propose an Anisotropic Spherical Gaussian (ASG)-based LDL approach for facial pose estimation. In particular, our approach adopts the spherical Gaussian distribution on a unit sphere which constantly generates unbiased expectation. Meanwhile, we introduce a new loss function that allows the network to learn the distribution parameter for each learning sample flexibly. Extensive experimental results show that our method sets new state-of-the-art records on AFLW2000 and BIWI datasets.
As AI technology is increasingly applied to high-impact, high-risk domains, there have been a number of new methods aimed at making AI models more human interpretable. Despite the recent growth of interpretability work, there is a lack of systematic evaluation of proposed techniques. In this work, we introduce HIVE (Human Interpretability of Visual Explanations), a novel human evaluation framework that assesses the utility of explanations to human users in AI-assisted decision making scenarios, and enables falsifiable hypothesis testing, cross-method comparison, and human-centered evaluation of visual interpretability methods. To the best of our knowledge, this is the first work of its kind. Using HIVE, we conduct IRB-approved human studies with nearly 1000 participants and evaluate four methods that represent the diversity of computer vision interpretability works: GradCAM, BagNet, ProtoPNet, and ProtoTree. Our results suggest that explanations engender human trust, even for incorrect predictions, yet are not distinct enough for users to distinguish between correct and incorrect predictions. We open-source HIVE to enable future studies and encourage more human-centered approaches to interpretability research.
Deep neural network quantization with adaptive bitwidths has gained increasing attention due to the ease of model deployment on various platforms with different resource budgets. In this paper, we propose a meta-learning approach to achieve this goal. Specifically, we propose MEBQAT, a simple yet effective way of bitwidth-adaptive quantization aware training (QAT) where meta-learning is effectively combined with QAT by redefining meta-learning tasks to incorporate bitwidths. After being deployed on a platform, MEBQAT allows the (meta-)trained model to be quantized to any candidate bitwidth then helps to conduct inference without much accuracy drop from quantization. Moreover, with a few-shot learning scenario, MEBQAT can also adapt a model to any bitwidth as well as any unseen target classes by adding conventional optimization or metric-based meta-learning. We design variants of MEBQAT to support both (1) a bitwidth-adaptive quantization scenario and (2) a new few-shot learning scenario where both quantization bitwidths and target classes are jointly adapted. We experimentally demonstrate their validity in multiple QAT schemes. By comparing their performance to (bitwidth-dedicated) QAT, existing bitwidth adaptive QAT and vanilla meta-learning, we find that merging bitwidths into meta-learning tasks achieves a higher level of robustness.
Continual learning (CL) requires a model to continually learn new tasks with incremental available information while retaining previous knowledge. Despite the numerous previous approaches to CL, most of them still suffer forgetting, expensive memory cost, or lack sufficient theoretical understanding. While different CL training regimes have been extensively studied empirically, insufficient attention has been paid to the underlying theory. In this paper, we establish a probabilistic framework to analyze information flow through layers in networks for sequential tasks and its impact on learning performance. Our objective is to optimize the information preservation between layers while learning new tasks. This manages task-specific knowledge passing throughout the layers while maintaining model performance on previous tasks. Our analysis provides novel insights into information adaptation within the layers during incremental task learning. We provide empirical evidence and practically highlight the performance improvement across multiple tasks. Code is available at https://github.com/Sekeh-Lab/InformationFlow-CL.
When optimizing sequentially incoming tasks, deep neural networks generally suffer from catastrophic forgetting due to their lack of ability to maintain knowledge from old tasks. This may lead to a significant performance drop of the previously learned tasks. To alleviate this problem, studies on continual learning have been conducted as a countermeasure. Nevertheless, it suffers from an increase in computational cost due to the expansion of the network size or a change in knowledge that is favorably linked to previous tasks. In this work, we propose a novel approach to differentiate helpful and harmful information for old tasks using a model search to learn a current task effectively. Given a new task, the proposed method discovers an underlying association knowledge from old tasks, which can provide additional support in acquiring the new task knowledge. In addition, by introducing a sensitivity measure to the loss of the current task from the associated tasks, we find cooperative relations between tasks while alleviating harmful interference. We apply the proposed approach to both task- and class-incremental scenarios in continual learning, using a wide range of datasets from small to large scales. Experimental results show that the proposed method outperforms a large variety of continual learning approaches for the experiments while effectively alleviating catastrophic forgetting.
Vision transformers (ViTs) have become the popular structures and outperformed convolutional neural networks (CNNs) on various vision tasks. However, such powerful transformers bring a huge computation burden, because of the exhausting token-to-token comparison. The previous works focus on dropping insignificant tokens to reduce the computational cost of ViTs. But when the dropping ratio increases, this hard manner will inevitably discard the vital tokens, which limits its efficiency. To solve the issue, we propose a generic self-slimmed learning approach for vanilla ViTs, namely SiT. Specifically, we first design a novel Token Slimming Module (TSM), which can boost the inference efficiency of ViTs by dynamic token aggregation. As a general method of token hard dropping, our TSM softly integrates redundant tokens into fewer informative ones. It can dynamically zoom visual attention without cutting off discriminative token relations in the images, even with a high slimming ratio. Furthermore, we introduce a concise Feature Recalibration Distillation (FRD) framework, wherein we design a reverse version of TSM (RTSM) to recalibrate the unstructured token in a flexible auto-encoder manner. Due to the similar structure between teacher and student, our FRD can effectively leverage structure knowledge for better convergence. Finally, we conduct extensive experiments to evaluate our SiT. It demonstrates that our method can speed up ViTs by 1.7x with negligible accuracy drop, and even speed up ViTs by 3.6x while maintaining 97% of their performance. Surprisingly, by simply arming LV-ViT with our SiT, we achieve new state-of-the-art performance on ImageNet. Code is available at https://github.com/Sense-X/SiT.
Modern iterations of deep learning models contain millions (billions) of unique parameters--each represented by a $b$-bit number. Popular attempts at compressing neural networks (such as pruning and quantisation) have shown that many of the parameters are superfluous, which we can remove (pruning) or express with $b’ < b$ bits (quantisation) without hindering performance. Here we look to go much further in minimising the information content of networks. Rather than a channel or layer-wise encoding, we look to lossless whole-network quantisation to minimise the entropy and number of unique parameters in a network. We propose a new method, which we call Weight Fixing Networks (WFN) that we design to realise four model outcome objectives: i) very few unique weights, ii) low-entropy weight encodings, iii) unique weight values which are amenable to energy-saving versions of hardware multiplication, and iv) lossless task-performance. Some of these goals are conflicting. To best balance these conflicts, we combine a few novel (and some well-trodden) tricks; a novel regularisation term, (i, ii) a view of clustering cost as relative distance change (i, ii, iv), and a focus on whole-network re-use of weights (i, iii). Our Imagenet experiments demonstrate lossless compression using 56x fewer unique weights and a 1.9x lower weight-space entropy than SOTA quantisation approaches.
This paper studies the Binary Neural Networks (BNNs) in which weights and activations are both binarized into 1-bit values, thus greatly reducing the memory usage and computational complexity. Since the modern deep neural networks are of sophisticated design with complex architecture for the accuracy reason, the diversity on distributions of weights and activations is very high. Therefore, the conventional sign function cannot be well used for effectively binarizing full precision values in BNNs. To this end, we present a simple yet effective approach called AdaBin to adaptively obtain the optimal binary sets {b_1, b_2} (b_1, b_2 belong to R) of weights and activations for each layer instead of a fixed set (i.e., {-1, +1}). In this way, the proposed method can better fit different distributions and increase the representation ability of binarized features. In practice, we use the center position and distance of 1-bit values to define a new binary quantization function. For the weights, we propose an equalization method to align the symmetrical center of binary distribution to real-valued distribution, and minimize the Kullback-Leibler divergence of them. Meanwhile, we introduce a gradient-based optimization method to get these two parameters for activations, which are jointly trained in an end-to-end manner. Experimental results on benchmark models and datasets demonstrate that the proposed AdaBin is able to achieve state-of-the-art performance. For instance, we obtain a 66.4% Top-1 accuracy on the ImageNet using ResNet-18 architecture, and a 69.4 mAP on PASCAL VOC using SSD300.
Self-attention based models such as vision transformers (ViTs) have emerged as a very competitive architecture alternative to convolutional neural networks (CNNs) in computer vision. Despite increasingly stronger variants with ever-higher recognition accuracies, due to the quadratic complexity of self-attention, existing ViTs are typically demanding in computation and model size. Although several successful design choices (e.g., the convolutions and hierarchical multi-stage structure) of prior CNNs have been reintroduced into recent ViTs, they are still not sufficient to meet the limited resource requirements of mobile devices. This motivates a very recent attempt to develop light ViTs based on the state-of-the-art MobileNet-v2, but still leaves a performance gap behind. In this work, pushing further along this under-studied direction we introduce EdgeViTs, a new family of light-weight ViTs that, for the first time, enable attention-based vision models to compete with the best light-weight CNNs in the tradeoff between accuracy and on-device efficiency. This is realized by introducing a highly cost-effective local-global-local (LGL) information exchange bottleneck based on optimal integration of self-attention and convolutions. For device-dedicated evaluation, rather than relying on inaccurate proxies like the number of FLOPs or parameters, we adopt a practical approach of focusing directly on on-device latency and, for the first time, energy efficiency. Extensive experiments on image classification, object detection, and semantic segmentation validate the high efficiency of our EdgeViTs when compared to the state-of-the-art efficient CNNs and ViTs in terms of accuracy-efficiency tradeoff on mobile hardware. Specifically, we show that our models are Pareto-optimal when both accuracy-latency and accuracy-energy tradeoffs are considered, achieving strict dominance over other ViTs in almost all cases and competing with the most efficient CNNs.
Knowledge Distillation (KD) transfers the knowledge from a high-capacity teacher network to strengthen a smaller student. Existing methods focus on excavating the knowledge hints and transferring the whole knowledge to the student. However, the knowledge redundancy arises since the knowledge shows different values to the student at different learning stages. In this paper, we propose Knowledge Condensation Distillation (KCD). Specifically, the knowledge value on each sample is dynamically estimated, based on which an Expectation-Maximization (EM) framework is forged to iteratively condense a compact knowledge set from the teacher to guide the student learning. Our approach is easy to build on top of the off-the-shelf KD methods, with no extra training parameters and negligible computation overhead. Thus, it presents one new perspective for KD, in which the student that actively identifies teacher’s knowledge in line with its aptitude can learn to learn more effectively and efficiently. Experiments on standard benchmarks manifest that the proposed KCD can well boost the performance of student model with even higher distillation efficiency. Code is available at https://github.com/dzy3/KCD.
Current object detection systems and benchmarks typically handle a limited number of categories, up to about a thousand categories. This paper scales the number of categories for object detection systems and benchmarks up to 21,000, by leveraging existing object detection and image classification data. Unlike previous efforts that usually transfer knowledge from base detectors to image classification data, we propose to rely more on a reverse information flow from a base image classifier to object detection data. In this framework, the large-vocabulary classification capability is first learnt thoroughly using only the image classification data. In this step, the image classification problem is reformulated as a special configuration of object detection that treats the entire image as a special RoI. Then, a simple multi-task learning approach is used to join the image classification and object detection data, with the backbone and the RoI classification branch shared between two tasks. This two-stage approach, though very simple without a sophisticated process such as multi-instance learning (MIL) to generate pseudo labels for object proposals on the image classification data, performs rather strongly that it surpasses previous large-vocabulary object detection systems on a standard evaluation protocol of tailored LVIS. Considering that the tailored LVIS evaluation only accounts for a few hundred novel object categories, we present a new evaluation benchmark that assesses the detection of all 21,841 object classes in the ImageNet-21K dataset. The baseline approach and evaluation benchmark will be publicly available at https://github.com/SwinTransformer/Simple-21K-Detection. We hope these would ease future research on large-vocabulary object detection.
What constitutes an object? This has been a long-standing question in computer vision. Towards this goal, numerous learning-free and learning-based approaches have been developed to score objectness. However, they generally do not scale well across new domains and for unseen objects. In this paper, we advocate that existing methods lack a top-down supervision signal governed by human-understandable semantics. For the first time in literature, we demonstrate that Multi-modal Vision Transformers (MViT) trained with aligned image-text pairs can effectively bridge this gap. Our extensive experiments across various domains and novel objects show the state-of-the-art performance of MViTs to localize generic objects in images. Based on the observation that existing MViTs do not include multi-scale feature processing and usually require longer training schedules, we develop an efficient and flexible MViT architecture using multi-scale and deformable self-attention. We show the significance of MViT proposals in a diverse range of applications including open-world object detection, salient and camouflage object detection, supervised and self-supervised detection tasks. Further, MViTs can adaptively generate proposals given a specific language query and thus offer enhanced interactability. Code: https://git.io/J1HPY.
We present ObjectBox, a novel single-stage anchor-free and highly generalizable object detection approach. As opposed to both existing anchor-based and anchor-free detectors, which are more biased toward specific object scales in their label assignments, we use only object center locations as positive samples and treat all objects equally in different feature levels regardless of the objects’ sizes or shapes. Specifically, our label assignment strategy considers the object center locations as shape- and size-agnostic anchors in an anchor-free fashion, and allows learning to occur at all scales for every object. To support this, we define new regression targets as the distances from two corners of the center cell location to the four sides of the bounding box. Moreover, to handle scale-variant objects, we propose a tailored IoU loss to deal with boxes with different sizes. As a result, our proposed object detector does not need any dataset-dependent hyperparameters to be tuned across datasets. We evaluate our method on MS-COCO 2017 and PASCAL VOC 2012 datasets, and compare our results to state-of-the-art methods. We observe that ObjectBox performs favorably in comparison to prior works. Furthermore, we perform rigorous ablation experiments to evaluate different components of our method. Our code is available at: https://github.com/MohsenZand/ObjectBox.
Absolute pose regressor (APR) networks are trained to estimate the pose of the camera given a captured image. They compute latent image representations from which the camera position and orientation are regressed. APRs provide a different tradeoff between localization accuracy, runtime, and memory, compared to structure-based localization schemes that provide state-of-the-art accuracy. In this work, we introduce Camera Pose Auto-Encoders (PAEs), multilayer perceptrons that are trained via a Teacher-Student approach to encode camera poses using APRs as their teachers. We show that the resulting latent pose representations can closely reproduce APR performance and demonstrate their effectiveness for related tasks. Specifically, we propose a light-weight test-time optimization in which the closest train poses are encoded and used to refine camera position estimation. This procedure achieves a new state-of-the-art position accuracy for APRs, on both the CambridgeLandmarks and 7Scenes benchmarks. We also show that train images can be reconstructed from the learned pose encoding, paving the way for integrating visual information from the train set at a low memory cost. Our code and pre-trained models are available at https://github.com/yolish/camera-pose-auto-encoders.
Light has many properties that vision sensors can passively measure. Colour-band separated wavelength and intensity are arguably the most commonly used for monocular 6D object pose estimation. This paper explores how complementary polarisation information, i.e. the orientation of light wave oscillations, influences the accuracy of pose predictions. A hybrid model that leverages physical priors jointly with a data-driven learning strategy is designed and carefully tested on objects with different levels of photometric complexity. Our design significantly improves the pose accuracy compared to state-of-the-art photometric approaches and enables object pose estimation for highly reflective and transparent objects. A new multi-modal instance-level 6D object pose dataset with highly accurate pose annotations for multiple objects with varying photometric complexity is introduced as a benchmark.
Detecting tiny objects is one of the main obstacles hindering the development of object detection. The performance of generic object detectors tends to drastically deteriorate on tiny object detection tasks. In this paper, we point out that either box prior in the anchor-based detector or point prior in the anchor-free detector is sub-optimal for tiny objects. Our key observation is that the current anchor-based or anchor-free label assignment paradigms will incur many outlier tiny-sized ground truth samples, leading to detectors imposing less focus on the tiny objects. To this end, we propose a Gaussian Receptive Field based Label Assignment (RFLA) strategy for tiny object detection. Specifically, RFLA first utilizes the prior information that the feature receptive field follows Gaussian distribution. Then, instead of assigning samples with IoU or center sampling strategy, a new Receptive Field Distance (RFD) is proposed to directly measure the similarity between the Gaussian receptive field and ground truth. Considering that the IoU-threshold based and center sampling strategy are skewed to large objects, we further design a Hierarchical Label Assignment (HLA) module based on RFD to achieve balanced learning for tiny objects. Extensive experiments on four datasets demonstrate the effectiveness of the proposed methods. Especially, our approach outperforms the state-of-the-art competitors with 4.0 AP points on the AI-TOD dataset. Codes are available at https://github.com/Chasel-Tsui/mmdet-rfla.
BEVFormer: Learning Bird’s-Eye-View Representation from Multi-Camera Images via Spatiotemporal Transformers
PDF ↗3D visual perception tasks, including 3D detection and map segmentation based on multi-camera images, are essential for autonomous driving systems. In this work, we present a new framework termed BEVFormer, which learns unified BEV representations with spatiotemporal transformers to support multiple autonomous driving perception tasks. In a nutshell, BEVFormer exploits both spatial and temporal information by interacting with spatial and temporal space through predefined grid-shaped BEV queries. To aggregate spatial information, we design spatial cross-attention that each BEV query extracts the spatial features from the regions of interest across camera views. For temporal information, we propose temporal self-attention to recurrently fuse the history BEV information. Our approach achieves the new state-of-the-art 56.9% in terms of NDS metric on the nuScenes test set, which is 9.0 points higher than previous best arts and on par with the performance of LiDAR-based baselines. We further show that BEVFormer remarkably improves the accuracy of velocity estimation and recall of objects under low visibility conditions. The code will be released at https://github.com/zhiqi-li/BEVFormer.