CVPR 2018 Papers

Accurate identification and localization of abnormalities from radiology images play an integral part in clinical diagnosis and treatment planning. Building a highly accurate prediction model for these tasks usually requires a large number of images manually annotated with labels and finding sites of abnormalities. In reality, however, such annotated data are expensive to acquire, especially the ones with location annotations. We need methods that can work well with only a small amount of location annotations. To address this challenge, we present a unified approach that simultaneously performs disease identification and localization through the same underlying model for all images.We demonstrate that our approach can effectively leverage both class information as well as limited location annotation, and significantly outperforms the comparative reference baseline in both classification and localization tasks.

While it is nearly effortless for humans to quickly assess the perceptual similarity between two images, the underlying processes are thought to be quite complex. Despite this, the most widely used perceptual metrics today, such as PSNR and SSIM, are simple, shallow functions, and fail to account for many nuances of human perception. Recently, the deep learning community has found that features of the VGG network trained on ImageNet classification has been remarkably useful as a training loss for image synthesis. But how perceptual are these so-called ``perceptual losses"? What elements are critical for their success? To answer these questions, we introduce a new dataset of human perceptual similarity judgments. We systematically evaluate deep features across different architectures and tasks and compare them with classic metrics. We find that deep features outperform all previous metrics by large margins on our dataset. More surprisingly, this result is not restricted to ImageNet-trained VGG features, but holds across different deep architectures and levels of supervision (supervised, self-supervised, or even unsupervised). Our results suggest that perceptual similarity is an emergent property shared across deep visual representations.

Deep learning methods have become the de-facto standard for challenging image processing tasks such as image classification. One major hurdle of deep learning approaches is that large sets of labeled data are necessary, which can be prohibitively costly to obtain, particularly in medical image diagnosis applications. Active learning techniques can alleviate this labeling effort. In this paper we investigate some recently proposed methods for active learning with high-dimensional data and convolutional neural network classifiers. We compare ensemble-based methods against Monte-Carlo Dropout and geometric approaches. We find that ensembles perform better and lead to more calibrated predictive uncertainties, which are the basis for many active learning algorithms. To investigate why Monte-Carlo Dropout uncertainties perform worse, we explore potential differences in isolation in a series of experiments. We show results for MNIST and CIFAR-10, on which we achieve a test set accuracy of $90 %$ with roughly 12,200 labeled images, and initial results on ImageNet. Additionally, we show results on a large, highly class-imbalanced diabetic retinopathy dataset. We observe that the ensemble-based active learning effectively counteracts this imbalance during acquisition.

Image restoration algorithms are typically evaluated by some distortion measure (e.g. PSNR, SSIM, IFC, VIF) or by human opinion scores that quantify perceived perceptual quality. In this paper, we prove mathematically that distortion and perceptual quality are at odds with each other. Specifically, we study the optimal probability for correctly discriminating the outputs of an image restoration algorithm from real images. We show that as the mean distortion decreases, this probability must increase (indicating worse perceptual quality). As opposed to the common belief, this result holds true for any distortion measure, and is not only a problem of the PSNR or SSIM criteria. However, as we show experimentally, for some measures it is less severe (e.g. distance between VGG features). We also show that generative-adversarial-nets (GANs) provide a principled way to approach the perception-distortion bound. This constitutes theoretical support to their observed success in low-level vision tasks. Based on our analysis, we propose a new methodology for evaluating image restoration methods, and use it to perform an extensive comparison between recent super-resolution algorithms.

The Jaccard index, also referred to as the intersection-over-union score, is commonly employed in the evaluation of image segmentation results given its perceptual qualities, scale invariance - which lends appropriate relevance to small objects, and appropriate counting of false negatives, in comparison to per-pixel losses. We present a method for direct optimization of the mean intersection-over-union loss in neural networks, in the context of semantic image segmentation, based on the convex Lovász extension of submodular losses. The loss is shown to perform better with respect to the Jaccard index measure than the traditionally used cross-entropy loss. We show quantitative and qualitative differences between optimizing the Jaccard index per image versus optimizing the Jaccard index taken over an entire dataset. We evaluate the impact of our method in a semantic segmentation pipeline and show substantially improved intersection-over-union segmentation scores on the Pascal VOC and Cityscapes datasets using state-of-the-art deep learning segmentation architectures.

Existing image classification datasets used in computer vision tend to have a uniform distribution of images across object categories. In contrast, the natural world is heavily imbalanced, as some species are more abundant and easier to photograph than others. To encourage further progress in challenging real world conditions we present the iNaturalist species classification and detection dataset, consisting of 859,000 images from over 5,000 different species of plants and animals. It features visually similar species, captured in a wide variety of situations, from all over the world. Images were collected with different camera types, have varying image quality, feature a large class imbalance, and have been verified by multiple citizen scientists. We discuss the collection of the dataset and present extensive baseline experiments using state-of-the-art computer vision classification and detection models. Results show that current non-ensemble based methods achieve only 67% top one classification accuracy, illustrating the difficulty of the dataset. Specifically, we observe poor results for classes with small numbers of training examples suggesting more attention is needed in low-shot learning.

Traditional methods of motion segmentation use powerful geometric constraints to understand motion, but fail to leverage the semantics of high-level image understanding. Modern CNN methods of motion analysis, on the other hand, excel at identifying well-known structures, but may not precisely characterize well-known geometric constraints. In this work, we build a new statistical model of rigid motion flow based on classical perspective projection constraints. We then combine piecewise rigid motions into complex deformable and articulated objects, guided by semantic segmentation from CNNs and a second ``object-level" statistical model. This combination of classical geometric knowledge combined with the pattern recognition abilities of CNNs yields excellent performance on a wide range of motion segmentation benchmarks, from complex geometric scenes to camouflaged animals.

In this paper, we investigate deep image synthesis guided by sketch, color, and texture. Previous image synthesis methods can be controlled by sketch and color strokes but we are the first to examine texture control. We allow a user to place a texture patch on a sketch at arbitrary locations and scales to control the desired output texture. Our generative network learns to synthesize objects consistent with these texture suggestions. To achieve this, we develop a local texture loss in addition to adversarial and content loss to train the generative network. We conduct experiments using sketches generated from real images and textures sampled from a separate texture database and results show that our proposed algorithm is able to generate plausible images that are faithful to user controls. Ablation studies show that our proposed pipeline can generate more realistic images than adapting existing methods directly.

Acquiring realistic texture details for 3D models is important in 3D reconstruction. However, the existence of geometric errors, caused by noisy RGB-D sensor data, always makes the color images cannot be accurately aligned onto reconstructed 3D models. In this paper, we propose a global-to-local correction strategy to obtain more desired texture mapping results. Our algorithm first adaptively selects an optimal image for each face of the 3D model, which can effectively remove blurring and ghost artifacts produced by multiple image blending. We then adopt a non-rigid global-to-local correction step to reduce the seaming effect between textures. This can effectively compensate for the texture and the geometric misalignment caused by camera pose drift and geometric errors. We evaluate the proposed algorithm in a range of complex scenes and demonstrate its effective performance in generating seamless high fidelity textures for 3D models.

Textbook Question Answering (TQA) is a task to choose the most proper answers by reading a multi-modal context of abundant essays and images. TQA serves as a favorable test bed for visual and textual reasoning. However, most of the current methods are incapable of reasoning over the long contexts and images. To address this issue, we propose a novel approach of Instructor Guidance with Memory Networks (IGMN) which conducts the TQA task by finding contradictions between the candidate answers and their corresponding context. We build the Contradiction Entity-Relationship Graph (CERG) to extend the passage-level multi-modal contradictions to an essay level. The machine thus performs as an instructor to extract the essay-level contradictions as the Guidance. Afterwards, we exploit the memory networks to capture the information in the Guidance, and use the attention mechanisms to jointly reason over the global features of the multi-modal input. Extensive experiments demonstrate that our method outperforms the state-of-the-arts on the TQA dataset. The source code is available at https://github.com/freerailway/igmn.

Visual relationships provide higher-level information of objects and their relations in an image – this enables a semantic understanding of the scene and helps downstream applications. Given a set of localized objects in some training data, visual relationship detection seeks to detect the most likely “relationship” between objects in a given image. While the specific objects may be well represented in training data, their relationships may still be infrequent. The empirical distribution obtained from seeing these relationships in a dataset does not model the underlying distribution well — a serious issue for most learning methods. In this work, we start from a simple multi-relational learning model, which in principle, offers a rich formalization for deriving a strong prior for learning visual relationships. While the inference problem for deriving the regularizer is challenging, our main technical contribution is to show how adapting recent results in numerical linear algebra lead to efficient algorithms for a factorization scheme that yields highly informative priors. The factorization provides sample size bounds for inference (under mild conditions) for the underlying [[object, predicate, object]] relationship learning task on its own and surprisingly outperforms (in some cases) existing methods even without utilizing visual features. Then, when integrated with an end to-end architecture for visual relationship detection leveraging image data, we substantially improve the state-of-the-art.

Action recognition in still images has been recently promoted by deep learning. However, the success of these deep models heavily depends on huge amount of training images for various action categories, which may not be available in practice. Alternatively, humans can classify new action categories after seeing few images, since we may not only compare appearance similarities between images on hand, but also attempt to recall importance motion cues from relevant action videos in our memory. To mimic this capacity, we propose a novel Hybrid Video Memory (HVM) machine, which can hallucinate temporal features of still images from video memory, in order to boost action recognition with few still images. First, we design a temporal memory module consisting of temporal hallucinating and predicting. Temporal hallucinating can generate temporal features of still images in an unsupervised manner. Hence, it can be flexibly used in realistic scenarios, where image and video categories may not be consistent. Temporal predicting can effectively infer action categories for query image, by integrating temporal features of training images and videos within a domain-adaptation manner. Second, we design a spatial memory module for spatial predicting. As spatial and temporal features are complementary to represent different actions, we apply spatial-temporal prediction fusion to further boost performance. Finally, we design a video selection module to select strongly-relevant videos as memory. In this case, we can balance the number of images and videos to reduce prediction bias as well as preserve computation efficiency. To show the effectiveness, we conduct extensive experiments on three challenging data sets, where our HVM outperforms a number of recent approaches by temporal hallucinating from video memory.

This paper is about temporal segmentation of human actions in videos. We introduce a new model -- temporal deformable residual network (TDRN) -- aimed at analyzing video intervals at multiple temporal scales for labeling video frames. Our TDRN computes two parallel temporal streams: i) Residual stream that analyzes video information at its full temporal resolution, and ii) Pooling/unpooling stream that captures long-range video information at different scales. The former facilitates local, fine-scale action segmentation, and the latter uses multiscale context for improving accuracy of frame classification. These two streams are computed by a set of temporal residual modules with deformable convolutions, and fused by temporal residuals at the full video resolution. Our evaluation on the University of Dundee 50 Salads, Georgia Tech Egocentric Activities, and JHU-ISI Gesture and Skill Assessment Working Set demonstrates that TDRN outperforms the state of the art in frame-wise segmentation accuracy, segmental edit score, and segmental overlap F1 score.

Weakly supervised learning with only coarse labels can obtain visual explanations of deep neural network such as attention maps by back-propagating gradients. These attention maps are then available as priors for tasks such as object localization and semantic segmentation. In one common framework we address three shortcomings of previous approaches in modeling such attention maps: We (1) make attention maps an explicit and natural component of the end-to-end training for the first time, (2) provide self-guidance directly on these maps by exploring supervision from the network itself to improve them, and (3) seamlessly bridge the gap between using weak and extra supervision if available. Despite its simplicity, experiments on the semantic segmentation task demonstrate the effectiveness of our methods. We clearly surpass the state-of-the-art on PASCAL VOC 2012 test and val. sets. Besides, the proposed framework provides a way not only explaining the focus of the learner but also feeding back with direct guidance towards specific tasks. Under mild assumptions our method can also be understood as a plug-in to existing weakly supervised learners to improve their generalization performance.

We study the problem of computer-assisted teaching with explanations. Conventional approaches for machine teaching typically only provide feedback at the instance level e.g., the category or label of the instance. However, it is intuitive that clear explanations from a knowledgeable teacher can significantly improve a student's ability to learn a new concept. To address these existing limitations, we propose a teaching framework that provides interpretable explanations as feedback and models how the learner incorporates this additional information. In the case of images, we show that we can automatically generate explanations that highlight the parts of the image that are responsible for the class label. Experiments on human learners illustrate that, on average, participants achieve better test set performance on challenging categorization tasks when taught with our interpretable approach compared to existing methods.

Do visual tasks have a relationship, or are they unrelated? For instance, could having surface normals simplify estimating the depth of an image? Intuition answers these questions positively, implying existence of a structure among visual tasks. Knowing this structure has notable uses; it is the concept underlying transfer learning and, for example, can provide a principled way for reusing supervision among related tasks, finding what tasks transfer well to an arbitrary target task, or solving many tasks in one system without piling up the complexity. This paper proposes a fully computational approach for finding the structure of the space of visual tasks. This is done via a sampled dictionary of twenty six 2D, 2.5D, 3D, and semantic tasks, and modeling their (1st and higher order) transfer dependencies in a latent space. The product can be viewed as a computational taxonomic map for task transfer learning. We study the consequences of this structure, e.g. the nontrivial emerged relationships, and exploit them to reduce the demand for labeled data. For example, we show that the total number of labeled datapoints needed for solving a set of 10 tasks can be reduced by roughly 2/3 while keeping the performance nearly the same. Users can employ a provided Binary Integer Programming solver that leverages the taxonomy to find efficient supervision policies for their own use cases.

We present an approach to semantic scene analysis using deep convolutional networks. Our approach is based on tangent convolutions - a new construction for convolutional networks on 3D data. In contrast to volumetric approaches, our method operates directly on surface geometry. Crucially, the construction is applicable to unstructured point clouds and other noisy real-world data. We show that tangent convolutions can be evaluated efficiently on large-scale point clouds with millions of points. Using tangent convolutions, we design a deep fully-convolutional network for semantic segmentation of 3D point clouds, and apply it to challenging real-world datasets of indoor and outdoor 3D environments. Experimental results show that the presented approach outperforms other recent deep network constructions in detailed analysis of large 3D scenes.

We propose a novel method for discovering shape regions that strongly correlate with user-prescribed tags. For example, given a collection of chairs tagged as either "has armrest" or "lacks armrest", our system correctly highlights the armrest regions as the main distinctive parts between the two chair types. To obtain point-wise predictions from shape-wise tags we develop a novel neural network architecture that is trained with tag classification loss, but is designed to rely on segmentation to predict the tag. Our network is inspired by U-Net, but we replicate shallow U structures several times with new skip connections and pooling layers, and call the resulting architecture "WU-Net". We test our method on segmentation benchmarks and show that even with weak supervision of whole shape tags, our method can infer meaningful semantic regions, without ever observing shape segmentations. Further, once trained, the model can process shapes for which the tag is entirely unknown. As a bonus, our architecture is directly operational under full supervision and performs strongly on standard benchmarks. We validate our method through experiments with many variant architectures and prior baselines, and demonstrate several applications.

In this work we propose a new automatic image annotation model, dubbed diverse and distinct image annotation (D2IA). The generative model D2IA is inspired by the ensemble of human annotations, which create semantically relevant, yet distinct and diverse tags. In D2IA, we generate a relevant and distinct tag subset, in which the tags are relevant to the image contents and semantically distinct to each other, using sequential sampling from a determinantal point process (DPP) model. Multiple such tag subsets that cover diverse semantic aspects or diverse semantic levels of the image contents are generated by randomly perturbing the DPP sampling process. We leverage a generative adversarial network (GAN) model to train D2IA. We perform extensive experiments including quantitative and qualitative comparisons, as well as human subject studies, on two benchmark datasets to demonstrate that the proposed model can produce more diverse and distinct tags than the state-of-the-arts.