CVPR 2018 Papers

Recent advances in video super-resolution have shown that convolutional neural networks combined with motion compensation are able to merge information from multiple low-resolution (LR) frames to generate high-quality images. Current state-of-the-art methods process a batch of LR frames to generate a single high-resolution (HR) frame and run this scheme in a sliding window fashion over the entire video, effectively treating the problem as a large number of separate multi-frame super-resolution tasks. This approach has two main weaknesses: 1) Each input frame is processed and warped multiple times, increasing the computational cost, and 2) each output frame is estimated independently conditioned on the input frames, limiting the system's ability to produce temporally consistent results. In this work, we propose an end-to-end trainable frame-recurrent video super-resolution framework that uses the previously inferred HR estimate to super-resolve the subsequent frame. This naturally encourages temporally consistent results and reduces the computational cost by warping only one image in each step. Furthermore, due to its recurrent nature, the proposed method has the ability to assimilate a large number of previous frames without increased computational demands. Extensive evaluations and comparisons with previous methods validate the strengths of our approach and demonstrate that the proposed framework is able to significantly outperform the current state of the art.

Adversarial attacks are known to succeed on classifiers, but it has been an open question whether more complex vision systems are vulnerable. In this paper, we study adversarial examples for vision and language models, which incorporate natural language understanding and complex structures such as attention, localization, and modular architectures. In particular, we investigate attacks on a dense captioning model and on two visual question answering (VQA) models. Our evaluation shows that we can generate adversarial examples with a high success rate (i.e., >90%) for these models. Our work sheds new light on understanding adversarial attacks on vision systems which have a language component and shows that attention, bounding box localization, and compositional internal structures are vulnerable to adversarial attacks. These observations will inform future work towards building effective defenses.

Recent deep networks that directly handle points in a point set, e.g., PointNet, have been state-of-the-art for supervised learning tasks on point clouds such as classification and segmentation. In this work, a novel end-to-end deep auto-encoder is proposed to address unsupervised learning challenges on point clouds. On the encoder side, a graph-based enhancement is enforced to promote local structures on top of PointNet. Then, a novel folding-based decoder deforms a canonical 2D grid onto the underlying 3D object surface of a point cloud, achieving low reconstruction errors even for objects with delicate structures. The proposed decoder only uses about 7% parameters of a decoder with fully-connected neural networks, yet leads to a more discriminative representation that achieves higher linear SVM classification accuracy than the benchmark. In addition, the proposed decoder structure is shown, in theory, to be a generic architecture that is able to reconstruct an arbitrary point cloud from a 2D grid. Our code is available at http://www.merl.com/research/license#FoldingNet

With the rise of misinformation spread via social media channels, enabled by the increasing automation and realism of image manipulation tools, image forensics is an increasingly relevant problem. Classic image forensic methods leverage low-level cues such as metadata, sensor noise fingerprints, and others that are easily fooled when the image is re-encoded upon upload to facebook, etc. This necessitates the use of higher-level physical and semantic cues that, once hard to estimate reliably in the wild, have become more effective due to the increasing power of computer vision. In particular, we detect manipulations introduced by artificial blurring of the image, which creates inconsistent photometric relationships between image intensity and various cues. We achieve 98% accuracy on the most challenging cases in a new dataset of blur manipulations, where the blur is geometrically correct and consistent with the scene's physical arrangement. Such manipulations are now easily generated, for instance, by smartphone cameras having hardware to measure depth, e.g. `Portrait Mode' of the iPhone7Plus. We also demonstrate good performance on a challenge dataset evaluating a wider range of manipulations in imagery representing `in the wild' conditions.

Recent insights on language and vision with neural networks have been successfully applied to simple single-image visual question answering. However, to tackle real-life question answering problems on multimedia collections such as personal photos, we have to look at whole collections with sequences of photos or videos. When answering questions from a large collection, a natural problem is to identify snippets to support the answer. In this paper, we describe a novel neural network called Focal Visual-Text Attention network (FVTA) for collective reasoning in visual question answering, where both visual and text sequence information such as images and text metadata are presented. FVTA introduces an end-to-end approach that makes use of a hierarchical process to dynamically determine what media and what time to focus on in the sequential data to answer the question. FVTA can not only answer the questions well but also provides the justifications which the system results are based upon to get the answers. FVTA achieves state-of-the-art performance on the MemexQA dataset and competitive results on the MovieQA dataset.

Image saliency detection has recently witnessed significant progress due to deep convolutional neural networks. However, extending state-of-the-art saliency detectors from image to video is challenging. The performance of salient object detection suffers from object or camera motion and the dramatic change of the appearance contrast in videos. In this paper, we present flow guided recurrent neural encoder(FGRNE), an accurate and end-to-end learning framework for video salient object detection. It works by enhancing the temporal coherence of the per-frame feature by exploiting both motion information in terms of optical flow and sequential feature evolution encoding in terms of LSTM networks. It can be considered as a universal framework to extend any FCN based static saliency detector to video salient object detection. Intensive experimental results verify the effectiveness of each part of FGRNE and confirm that our proposed method significantly outperforms state-of-the-art methods on the public benchmarks of DAVIS and FBMS.

We present FlipDial, a generative model for Visual Dialogue that simultaneously plays the role of both participants in a visually-grounded dialogue. Given context in the form of an image and an associated caption summarising the contents of the image, FlipDial learns both to answer questions and put forward questions, capable of generating entire sequences of dialogue (question-answer pairs) which are diverse and relevant to the image. To do this, FlipDial relies on a simple but surprisingly powerful idea: it uses convolutional neural networks (CNNs) to encode entire dialogues directly, implicitly capturing dialogue context, and conditional VAEs to learn the generative model. FlipDial outperforms the state-of-the-art model in the sequential answering task (1VD) on the VisDial dataset by 5 points in Mean Rank using the generated answers. We are the first to extend this paradigm to full two-way visual dialogue (2VD), where our model is capable of generating both questions and answers in sequence based on a visual input, for which we propose a set of novel evaluation measures and metrics.

We aim at estimating the fundamental matrix in two views from five correspondences of rotation invariant features obtained by e.g. the SIFT detector. The proposed minimal solver first estimates a homography from three correspondences assuming that they are co-planar and exploiting their rotational components. Then the fundamental matrix is obtained from the homography and two additional point pairs in general position. The proposed approach, combined with robust estimators like Graph-Cut RANSAC, is superior to other state-of-the-art algorithms both in terms of accuracy and number of iterations required. This is validated on synthesized data and 561 real image pairs. Moreover, the tests show that requiring three points on a plane is not too restrictive in urban environment and locally optimized robust estimators lead to accurate estimates even if the points are not entirely co-planar. As a potential application, we show that using the proposed method makes two-view multi-motion estimation more accurate.

In this work we study the use of 3D hand poses to recognize first-person dynamic hand actions interacting with 3D objects. Towards this goal, we collected RGB-D video sequences comprised of more than 100K frames of 45 daily hand action categories, involving 26 different objects in several hand configurations. To obtain hand pose annotations, we used our own mo-cap system that automatically infers the 3D location of each of the 21 joints of a hand model via 6 magnetic sensors and inverse kinematics. Additionally, we recorded the 6D object poses and provide 3D object models for a subset of hand-object interaction sequences. To the best of our knowledge, this is the first benchmark that enables the study of first-person hand actions with the use of 3D hand poses. We present an extensive experimental evaluation of RGB-D and pose-based action recognition by 18 baselines/state-of-the-art approaches. The impact of using appearance features, poses, and their combinations are measured, and the different training/testing protocols are evaluated. Finally, we assess how ready the 3D hand pose estimation field is when hands are severely occluded by objects in egocentric views and its influence on action recognition. From the results, we see clear benefits of using hand pose as a cue for action recognition compared to other data modalities. Our dataset and experiments can be of interest to communities of 3D hand pose estimation, 6D object pose, and robotics as well as action recognition.

Despite recent emergence of video caption methods, how to generate fine-grained video descriptions (i.e., long and detailed commentary about individual movements of multiple subjects as well as their frequent interactions) is far from being solved, which however has great applications such as automatic sports narrative. To this end, this work makes the following contributions. First, to facilitate this novel research of fine-grained video caption, we collected a novel dataset called Fine-grained Sports Narrative dataset (FSN) that contains 2K sports videos with ground-truth narratives from YouTube.com. Second, we develop a novel performance evaluation metric named Fine-grained Captioning Evaluation (FCE) to cope with this novel task. Considered as an extension of the widely used METEOR, it measures not only the linguistic performance but also whether the action details and their temporal orders are correctly described. Third, we propose a new framework for fine-grained sports narrative task. This network features three branches: 1) a spatio-temporal entity localization and role discovering sub-network; 2) a fine-grained action modeling sub-network for local skeleton motion description; and 3) a group relationship modeling sub-network to model interactions between players. We further fuse the features and decode them into long narratives by a hierarchically recurrent structure. Extensive experiments on the FSN dataset demonstrates the validity of the proposed framework for fine-grained video caption.

Face detection techniques have been developed for decades, and one of remaining open challenges is detecting small faces in unconstrained conditions. The reason is that tiny faces are often lacking detailed information and blurring. In this paper, we proposed an algorithm to directly generate a clear high-resolution face from a blurry small one by adopting a generative adversarial network (GAN). Toward this end, the basic GAN formulation achieves it by super-resolving and refining sequentially (e.g. SR-GAN and cycle-GAN). However, we design a novel network to address the problem of super-resolving and refining jointly. We also introduce new training losses to guide the generator network to recover fine details and to promote the discriminator network to distinguish real vs. fake and face vs. non-face simultaneously. Extensive experiments on the challenging dataset WIDER FACE demonstrate the effectiveness of our proposed method in restoring a clear high-resolution face from a blurry small one, and show that the detection performance outperforms other state-of-the-art methods.

Several works have proposed to learn a two-path neural network that maps images and texts, respectively, to a same shared Euclidean space where geometry captures useful semantic relationships. Such a multi-modal embedding can be trained and used for various tasks, notably image captioning. In the present work, we introduce a new architecture of this type, with a visual path that leverages recent space-aware pooling mechanisms. Combined with a textual path which is jointly trained from scratch, our semantic-visual embedding offers a versatile model. Once trained under the supervision of captioned images, it yields new state-of-the-art performance on cross-modal retrieval. It also allows the localization of new concepts from the embedding space into any input image, delivering state-of-the-art result on the visual grounding of phrases.

Grounding textual phrases in visual content with standalone image-sentence pairs is a challenging task. When we consider grounding in instructional videos, this problem becomes profoundly more complex: the latent temporal structure of instructional videos breaks independence assumptions and necessitates contextual understanding for resolving ambiguous visual-linguistic cues. Furthermore, dense annotations and video data scale mean supervised approaches are prohibitively costly. In this work, we propose to tackle this new task with a weakly-supervised framework for reference-aware visual grounding in instructional videos, where only the temporal alignment between the transcription and the video segment are available for supervision. We introduce the visually grounded action graph, a structured representation capturing the latent dependency between grounding and references in video. For optimization, we propose a new reference-aware multiple instance learning (RA-MIL) objective for weak supervision of grounding in videos. We evaluate our approach over unconstrained videos from YouCookII and RoboWatch, augmented with new reference-grounding test set annotations. We demonstrate that our jointly optimized, reference-aware approach simultaneously improves visual grounding, reference-resolution, and generalization to unseen instructional video categories.

We put forward a principled variational approach for up-sampling a single depth map to the resolution of the companion color image provided by an RGB-D sensor. We combine heterogeneous depth and color data in order to jointly solve the ill-posed depth super-resolution and shape-from-shading problems. The low-frequency geometric information necessary to disambiguate shape-from-shading is extracted from the low-resolution depth measurements and, symmetrically, the high-resolution photometric clues in the RGB image provide the high-frequency information required to disambiguate depth super-resolution.

In this paper, we are interested in the few-shot learning problem. In particular, we focus on a challenging scenario where the number of categories is large and the number of examples per novel category is very limited, e.g. 1, 2, or 3. Motivated by the close relationship between the parameters and the activations in a neural network associated with the same category, we propose a novel method that can adapt a pre-trained neural network to novel categories by directly predicting the parameters from the activations. Zero training is required in adaptation to novel categories, and fast inference is realized by a single forward pass. We evaluate our method by doing few-shot image recognition on the ImageNet dataset, which achieves the state-of-the-art classification accuracy on novel categories by a significant margin while keeping comparable performance on the large-scale categories. We also test our method on the MiniImageNet dataset and it strongly outperforms the previous state-of-the-art methods.

We propose an inference procedure for deep convolutional neural networks (CNNs) when partial evidence is available. Our method consists of a general feedback-based propagation approach (feedback-prop) that boosts the prediction accuracy for an arbitrary set of unknown target labels when the values for a non-overlapping arbitrary set of target labels are known. We show that existing models trained in a multi-label or multi-task setting can readily take advantage of feedback-prop without any retraining or fine-tuning. Our feedback-prop inference procedure is general, simple, reliable, and works on different challenging visual recognition tasks. We present two variants of feedback-prop based on layer-wise and residual iterative updates. We experiment using several multi-task models and show that feedback-prop is effective in all of them. Our results unveil a previously unreported but interesting dynamic property of deep CNNs. We also present an associated technical approach that takes advantage of this property for inference under partial evidence in general visual recognition tasks.

Multi-Target Multi-Camera Tracking (MTMCT) tracks many people through video taken from several cameras. Person Re-Identification (Re-ID) retrieves from a gallery images of people similar to a person query image. We learn good features for both MTMCT and Re-ID with a convolutional neural network. Our contributions include an adaptive weighted triplet loss for training and a new technique for hard-identity mining. Our method outperforms the state of the art both on the DukeMTMC benchmarks for tracking, and on the Market-1501 and DukeMTMC-ReID benchmarks for Re-ID. We examine the correlation between good Re-ID and good MTMCT scores, and perform ablation studies to elucidate the contributions of the main components of our system. Code is available.

Identifying small size images or small objects is a notoriously challenging problem, as discriminative representations are difficult to learn from the limited information contained in them with poor-quality appearance and unclear object structure. Existing research works usually increase the resolution of low-resolution image in the pixel space in order to provide better visual quality for human viewing. However, the improved performance of such methods is usually limited or even trivial in the case of very small image size (we will show it in this paper explicitly). In this paper, different from image super-resolution (ISR), we propose a novel super-resolution technique called feature super-resolution (FSR), which aims at enhancing the discriminatory power of small size image in order to provide high recognition precision for machine. To achieve this goal, we propose a new Feature Super-Resolution Generative Adversarial Network (FSR-GAN) model that transforms the raw poor features of small size images to highly discriminative ones by performing super-resolution in the feature space. Our FSR-GAN consists of two subnetworks: a feature generator network G and a feature discriminator network D. By training the G and the D networks in an alternative manner, we encourage the G network to discover the latent distribution correlations between small size and large size images and then use G to improve the representations of small images. Extensive experiment results on Oxford5K, Paris, Holidays, and Flick100k datasets demonstrate that the proposed FSR approach can effectively enhance the discriminatory ability of features. Even when the resolution of query images is reduced greatly, e.g., 1/64 original size, the query feature enhanced by our FSR approach achieves surprisingly high retrieval performance at different image resolutions and increases the retrieval precision by 25% compared to the raw query feature.

The problem of data augmentation in feature space is considered. A new architecture, denoted the FeATure TransfEr Network (FATTEN), is proposed for the modeling of feature trajectories induced by variations of object pose. This architecture exploits a parametrization of the pose manifold in terms of pose and appearance. This leads to a deep encoder/decoder network architecture, where the encoder factors into an appearance and a pose predictor. Unlike previous attempts at trajectory transfer, FATTEN can be efficiently trained end-to-end, with no need to train separate feature transfer functions. This is realized by supplying the decoder with information about a target pose and the use of a multi-task loss that penalizes category- and pose-mismatches. In result, FATTEN discourages discontinuous or non-smooth trajectories that fail to capture the structure of the pose manifold, and generalizes well on object recognition tasks involving large pose variation. Experimental results on the artificial ModelNet database show that it can successfully learn to map source features to target features of a desired pose, while preserving class identity. Most notably, by using feature space transfer for data augmentation (w.r.t. pose and depth) on SUN-RGBD objects, we demonstrate considerable performance improvements on one/few-shot object recognition in a transfer learning setup, compared to current state-of-the-art methods.