CVPR 2021 Papers

Reasoning the human-object interactions (HOI) is essential for deeper scene understanding, while object affordances (or functionalities) are of great importance for human to discover unseen HOIs with novel objects. Inspired by this, we introduce an affordance transfer learning approach to jointly detect HOIs with novel object and recognize affordances. Specifically, HOI representations can be decoupled into a combination of affordance and object representations, making it possible to compose novel interactions by combining affordance representations and novel object representations from additional images, i.e. transferring the affordance to novel objects. With the proposed affordance transfer learning, the model is also capable of inferring the affordances of novel objects from known affordance representations. The proposed method can thus be used to 1) improve the performance of HOI detection, especially for the HOIs with unseen objects; and 2) infer the affordances of novel objects. Experimental results on two datasets, HICO-DET and HOI-COCO (from V-COCO), demonstrate significant improvements over recent state-of-the-art methods for HOI detection and object affordance detection. Code is available at https://github.com/zhihou7/HOI-CL.

Temporal context is key to the recognition of expressions of emotion. Existing methods, that rely on recurrent or self-attention models to enforce temporal consistency, work on the feature level, ignoring the task-specific temporal dependencies, and fail to model context uncertainty. To alleviate these issues, we build upon the framework of Neural Processes to propose a method for apparent emotion recognition with three key novel components: (a) probabilistic contextual representation with a global latent variable model; (b) temporal context modelling using task-specific predictions in addition to features; and (c) smart temporal context selection. We validate our approach on four databases, two for Valence and Arousal estimation (SEWA and AffWild2), and two for Action Unit intensity estimation (DISFA and BP4D). Results show a consistent improvement over a series of strong baselines as well as over state-of-the-art methods.

We present Affect2MM, a learning method for time-series emotion prediction for multimedia content. Our goal is to automatically capture the varying emotions depicted by characters in real-life human-centric situations and behaviors. We use the ideas from emotion causation theories to computationally model and determine the emotional state evoked in clips of movies. Affect2MM explicitly models the temporal causality using attention-based methods and Granger causality. We use a variety of components like facial features of actors involved, scene understanding, visual aesthetics, action/situation description, and movie script to obtain an affective-rich representation to understand and perceive the scene. We use an LSTM-based learning model for emotion perception. To evaluate our method, we analyze and compare our performance on three datasets, SENDv1, MovieGraphs, and the LIRIS-ACCEDE dataset, and observe an average of 10-15% increase in the performance over SOTA methods for all three datasets.

Single domain generalization aims to learn a model that performs well on many unseen domains with only one domain data for training. Existing works focus on studying the adversarial domain augmentation (ADA) to improve the model's generalization capability. The impact on domain generalization from the statistics of normalization layers is still underinvestigated. In this paper, we propose a generic normalization approach, adaptive standardization and rescaling normalization (ASR-Norm), to complement the missing part in the previous works. ASR-Norm learns both the standardization and rescaling statistics via neural networks. This new form of normalization can be viewed as a generic form of traditional normalizations. When trained with ADA, the statistics in ASR-Norm are learned to be adaptive to the data coming from different domains, and hence improves the model generalization performance across domains, especially on the target domain with large discrepancy from the source domain. The experimental results show that ASR-Norm can bring consistent improvement to the state-of-the-art ADA approaches by 1.6%, 2.7%, and 6.3% averagely on the Digits, CIFAR-10-C, and PACS benchmarks, respectively. As a generic tool, the improvement introduced by ASR-Norm is agnostic to the choice of ADA methods.

Adversarial robustness has attracted extensive studies recently by revealing the vulnerability and intrinsic characteristics of deep networks. However, existing works on adversarial robustness mainly focus on balanced datasets, while real-world data usually exhibits a long-tailed distribution. To push adversarial robustness towards more realistic scenarios, in this work we investigate the adversarial vulnerability as well as defense under long-tailed distributions. In particular, we first reveal the negative impacts induced by imbalanced data on both recognition performance and adversarial robustness, uncovering the intrinsic challenges of this problem. We then perform a systematic study on existing long-tailed recognition methods in conjunction with the adversarial training framework. Several valuable observations are obtained: 1) natural accuracy is relatively easy to improve, 2) fake gain of robust accuracy exists under unreliable evaluation, and 3) boundary error limits the promotion of robustness. Inspired by these observations, we propose a clean yet effective framework, RoBal, which consists of two dedicated modules, a scale-invariant classifier and data re-balancing via both margin engineering at training stage and boundary adjustment during inference. Extensive experiments demonstrate the superiority of our approach over other state-of-the-art defense methods. To our best knowledge, we are the first to tackle adversarial robustness under long-tailed distributions, which we believe would be a significant step towards real-world robustness. Our code is available at: https://github.com/wutong16/Adversarial_Long-Tail.

Adversarial robustness corresponds to the susceptibility of deep neural networks to imperceptible perturbations made at test time. In the context of image tasks, many algorithms have been proposed to make neural networks robust to adversarial perturbations made to the input pixels. These perturbations are typically measured in an l_p norm. However, robustness often holds only for the specific attack used for training. In this work we extend the above setting to consider the problem of training of deep neural networks that can be made simultaneously robust to perturbations applied in multiple natural representations spaces. For the case of image data, examples include the standard pixel representation as well as the representation in the discrete cosine transform (DCT) basis. We design a theoretically sound algorithm with formal guarantees for the above problem. Furthermore, our guarantees also hold when the goal is to require robustness with respect to multiple l_p norm based attacks. We then derive an efficient practical implementation and demonstrate the effectiveness of our approach on standard datasets for image classification.

Though it is well known that the performance of deep neural networks (DNNs) degrades under certain light conditions, there exists no study on the threats of light beams emitted from some physical source as adversarial attacker on DNNs in a real-world scenario. In this work, we show by simply using a laser beam that DNNs are easily fooled. To this end, we propose a novel attack method called Adversarial Laser Beam (AdvLB), which enables manipulation of laser beam's physical parameters to perform adversarial attack. Experiments demonstrate the effectiveness of our proposed approach in both digital- and physical-settings. We further empirically analyze the evaluation results and reveal that the proposed laser beam attack may lead to some interesting prediction errors of the state-of-the-art DNNs. We envisage that the proposed AdvLB method enriches the current family of adversarial attacks and builds the foundation for future robustness studies for light.

Though machine learning algorithms are able to achieve pattern recognition from the correlation between data and labels, the presence of spurious features in the data decreases the robustness of these learned relationships with respect to varied testing environments. This is known as out-of-distribution (OoD) generalization problem. Recently, invariant risk minimization (IRM) attempts to tackle this issue by penalizing predictions based on the unstable spurious features in the data collected from different environments. However, similar to domain adaptation or domain generalization, a prevalent non-trivial limitation in these works is that the environment information is assigned by human specialists i.e. a priori or determined heuristically. However, an inappropriate group partitioning can dramatically deteriorate the OoD generalization and the process is expensive and time-consuming. To deal with this issue, we propose a novel theoretically principled min-max framework to iteratively construct a worst-case splitting, i.e. creating the most challenging environment splittings for the backbone learning paradigm (e.g. IRM) to learn the robust feature representation. We also design a differentiable training strategy to facilitate the feasible gradient-based computation. Numerical experiments show that our algorithmic framework has achieved superior and stable performance in various datasets, such as Colored MNIST and Punctuated Stanford Sentiment Treebank (SST). Furthermore, we also find our algorithm to be robust even to a strong data poisoning attack. To the best of our knowledge, this is one of the first to adopt differentiable environment splitting method to enable stable predictions across environments without environment index information, which achieves the state-of-the-art performance on datasets with strong spurious correlation, such as Colored MNIST.

Adversarial attacks play a critical role in understanding deep neural network predictions and improving their robustness. Existing attack methods aim to deceive convolutional neural network (CNN)-based classifiers by manipulating RGB images that are fed directly to the classifiers. However, these approaches typically neglect the influence of the camera optics and image processing pipeline (ISP) that produce the network inputs. ISPs transform RAW measurements to RGB images and traditionally are assumed to preserve adversarial patterns. In fact, these low-level pipelines can destroy, introduce or amplify adversarial patterns that can deceive a downstream detector. As a result, optimized patterns can become adversarial for the classifier after being transformed by a certain camera ISP or optical lens system but not for others. In this work, we examine and develop such an attack that deceives a specific camera ISP while leaving others intact, using the same downstream classifier. We frame this camera-specific attack as a multi-task optimization problem, relying on a differentiable approximation for the ISP itself. We validate the proposed method using recent state-of-the-art automotive hardware ISPs, achieving 92% fooling rate when attacking a specific ISP. We demonstrate physical optics attacks with 90% fooling rate for a specific camera lens.

In most existing learning systems, images are typically viewed as 2D pixel arrays. However, in another paradigm gaining popularity, a 2D image is represented as an implicit neural representation (INR) -- an MLP that predicts an RGB pixel value given its (x,y) coordinate. In this paper, we propose two novel architectural techniques for building INR-based image decoders: factorized multiplicative modulation and multi-scale INRs, and use them to build a state-of-the-art continuous image GAN. Previous attempts to adapt INRs for image generation were limited to MNIST-like datasets and do not scale to complex real-world data. Our proposed INR-GAN architecture improves the performance of continuous image generators by several times, greatly reducing the gap between continuous image GANs and pixel-based ones. Apart from that, we explore several exciting properties of the INR-based decoders, like out-of-the-box superresolution, meaningful image-space interpolation, accelerated inference of low-resolution images, an ability to extrapolate outside of image boundaries, and strong geometric prior. The project page is located at https://universome.github.io/inr-gan.

As self-driving systems become better, simulating scenarios where the autonomy stack may fail becomes more important. Traditionally, those scenarios are generated for a few scenes with respect to the planning module that takes ground-truth actor states as input. This does not scale and cannot identify all possible autonomy failures, such as perception failures due to occlusion. In this paper, we propose AdvSim, an adversarial framework to generate safety-critical scenarios for any LiDAR-based autonomy system. Given an initial traffic scenario, AdvSim modifies the actors' trajectories in a physically plausible manner and updates the LiDAR sensor data to match the perturbed world. Importantly, by simulating directly from sensor data, we obtain adversarial scenarios that are safety-critical for the full autonomy stack. Our experiments show that our approach is general and can identify thousands of semantically meaningful safety-critical scenarios for a wide range of modern self-driving systems. Furthermore, we show that the robustness and safety of these systems can be further improved by training them with scenarios generated by AdvSim.

This paper studies the single image super-resolution problem using adder neural networks (AdderNets). Compared with convolutional neural networks, AdderNets utilize additions to calculate the output features thus avoid massive energy consumptions of conventional multiplications. However, it is very hard to directly inherit the existing success of AdderNets on large-scale image classification to the image super-resolution task due to the different calculation paradigm. Specifically, the adder operation cannot easily learn the identity mapping, which is essential for image processing tasks. In addition, the functionality of high-pass filters cannot be ensured by AdderNets. To this end, we thoroughly analyze the relationship between an adder operation and the identity mapping and insert shortcuts to enhance the performance of SR models using adder networks. Then, we develop a learnable power activation for adjusting the feature distribution and refining details. Experiments conducted on several benchmark models and datasets demonstrate that, our image super-resolution models using AdderNets can achieve comparable performance and visual quality to that of their CNN baselines with an about 2.5x reduction on the energy consumption. The codes are available at: https://github.com/huawei-noah/AdderNet.

Generative adversarial network (GAN) has become one of the most important neural network models for classical unsupervised machine learning. A variety of discriminator loss functions have been developed to train GAN's discriminators and they all have a common structure: a sum of real and fake losses that only depends on the actual and generated data respectively. One challenge associated with an equally weighted sum of two losses is that the training may benefit one loss but harm the other, which we show causes instability and mode collapse. In this paper, we introduce a new family of discriminator loss functions that adopts a weighted sum of real and fake parts, which we call adaptive weighted loss functions or aw-loss functions. Using the gradients of the real and fake parts of the loss, we can adaptively choose weights to train a discriminator in the direction that benefits the GAN's stability. Our method can be potentially applied to any discriminator model with a loss that is a sum of the real and fake parts. Experiments validated the effectiveness of our loss functions on unconditional and conditional image generation tasks, improving the baseline results by a significant margin on CIFAR-10, STL-10, and CIFAR-100 datasets in Inception Scores (IS) and Frechet Inception Distance (FID) metrics.

Robust principal component analysis (RPCA) and its variants have gained wide applications in computer vision. However, these methods either involve manual adjustment of some parameters, or require the rank of a low-rank matrix to be known a prior. In this paper, an adaptive rank estimate based RPCA (ARE-RPCA) is proposed, which adaptively assigns weights on different singular values via rank estimation. More specifically, we study the characteristics of the low-rank matrix, and develop an improved Gerschgorin disk theorem to estimate the rank of the low-rank matrix accurately. Furthermore in view of the issue occurred in the Gerschgorin disk theorem that adjustment factor need to be manually pre-defined, an adaptive setting method, which greatly facilitates the practical implementation of the rank estimation, is presented. Then, the weights of singular values in the nuclear norm are updated adaptively based on iteratively estimated rank, and the resultant low-rank matrix is close to the target. Experimental results show that the proposed ARE-RPCA outperforms the state-of-the-art methods in various complex scenarios.

Prototype learning is extensively used for few-shot segmentation. Typically, a single prototype is obtained from the support feature by averaging the global object information. However, using one prototype to represent all the information may lead to ambiguities. In this paper, we propose two novel modules, named superpixel-guided clustering (SGC) and guided prototype allocation (GPA), for multiple prototype extraction and allocation. Specifically, SGC is a parameter-free and training-free approach, which extracts more representative prototypes by aggregating similar feature vectors, while GPA is able to select matched prototypes to provide more accurate guidance. By integrating the SGC and GPA together, we propose the Adaptive Superpixel-guided Network (ASGNet), which is a lightweight model and adapts to object scale and shape variation. In addition, our network can easily generalize to k-shot segmentation with substantial improvement and no additional computational cost. In particular, our evaluations on COCO demonstrate that ASGNet surpasses the state-of-the-art method by 5% in 5-shot segmentation.

Invariant approaches have been remarkably successful in tackling the problem of domain generalization, where the objective is to perform inference on data distributions different from those used in training. In our work, we investigate whether it is possible to leverage domain information from the unseen test samples themselves. We propose a domain-adaptive approach consisting of two steps: a) we first learn a discriminative domain embedding from unsupervised training examples, and b) use this domain embedding as supplementary information to build a domain-adaptive model, that takes both the input as well as its domain into account while making predictions. For unseen domains, our method simply uses few unlabelled test examples to construct the domain embedding. This enables adaptive classification on any unseen domain. Our approach achieves state-of-the-art performance on various domain generalization benchmarks. In addition, we introduce the first real-world, large-scale domain generalization benchmark, Geo-YFCC, containing 1.1M samples over 40 training, 7 validation and 15 test domains, orders of magnitude larger than prior work. We show that the existing approaches either do not scale to this dataset or underperform compared to the simple baseline of training a model on the union of data from all training domains. In contrast, our approach achieves a significant 1% improvement.

One-shot object detection tackles a challenging task that aims at identifying within a target image all object instances of the same class, implied by a query image patch. The main difficulty lies in the situation that the class label of the query patch and its respective examples are not available in the training data. Our main idea leverages the concept of language translation to boost metric-learning-based detection methods. Specifically, we emulate the language translation process to adaptively translate the feature of each object proposal to better correlate the given query feature for discriminating the class-similarity among the proposal-query pairs. To this end, we propose the Adaptive Image Transformer (AIT) module that deploys an attention-based encoder-decoder architecture to simultaneously explore intra-coder and inter-coder (i.e., each proposal-query pair) attention. The adaptive nature of our design turns out to be flexible and effective in addressing the one-shot learning scenario. With the informative attention cues, the proposed model excels in predicting the class-similarity between the target image proposals and the query image patch. Though conceptually simple, our model significantly outperforms a state-of-the-art technique, improving the unseen-class object classification from 63.8 mAP and 22.0 AP50 to 72.2 mAP and 24.3 AP50 on the PASCAL-VOC and MS-COCO benchmark datasets, respectively.

In this paper, we study the task of visual-text retrieval in the highly practical setting in which labelled visual data with paired text descriptions are available in one domain (the "source"), but only unlabelled visual data (without text descriptions) are available in the domain of interest (the "target"). We propose the ADAPTIVE CROSS-MODAL PROTOTYPES framework which seeks to enable target domain retrieval by learning cross-modal visual-text representations while minimising both uni-modal and cross-modal distribution shift across the source and target domains. Our approach is built upon two key ideas: first, we encode the inductive bias that the learned cross-modal representations should be compositional with respect to concepts in each modality--this is achieved through clustering pretrained uni-modal features across each domain and designing a careful regularisation scheme to preserve the resulting structure. Second, we employ mutual information maximisation between cross-modal representations in the source and target domains during learning--this provides a mechanism that preserves commonalities between the domains while discarding signal in each that cannot be inferred from the other. We showcase our approach for the task of cross-domain visual-text retrieval, outperforming existing approaches for both images and videos.

Style transfer between images is an artistic application of CNNs, where the 'style' of one image is transferred onto another image while preserving the latter's content. The state of the art in neural style transfer is based on Adaptive Instance Normalization (AdaIN), a technique that transfers the statistical properties of style features to a content image, and can transfer a large number of styles in real time. However, AdaIN is a global operation; thus local geometric structures in the style image are often ignored during the transfer. We propose Adaptive Convolutions (AdaConv), a generic extension of AdaIN, to allow for the simultaneous transfer of both statistical and structural styles in real time. Apart from style transfer, our method can also be readily extended to style-based image generation, and other tasks where AdaIN has already been adopted.