ICCV 2017 Papers

Deep learning has achieved great success in face recognition, however deep-learned features still have limited invariance to strong intra-personal variations such as large pose. It is observed that some facial attributes (e.g. eyebrow thickness, gender) are invariant to such variations. We present the first work to systematically explore how the fusion of face recognition feature (FRF) and facial attribute feature (FAF) can enhance face recognition performance in various challenging scenarios. Despite this helpfulness of FAF, in practice, we find the existing fusion methods cannot reliably improve the recognition performance. Thus, we develop a powerful tensor-based framework which formulates this fusion as a low-rank tensor optimisation problem. It is non-trivial to directly optimise this tensor due to the large number of parameters to optimise. To solve this problem, we establish a theoretical equivalence between tensor optimisation and a two-stream gated neural network. This equivalence allows tractable computation and the use of standard neural network optimisation tools, leading to an accurate and stable optimisation. Experimental results show the fused feature works better than individual features thus proving for the first time that facial attributes aid face recognition. We achieve state-of-the-art performance on databases such as MultiPIE, CASIA NIR-VIR2.0 and LFW.

Recognising semantic pedestrian attributes in surveillance images is a challenging task for computer vision, particularly when the imaging quality is poor with complex background clutter and uncontrolled viewing conditions, and the number of labelled training data is small. In this work, we formulate a Joint Recurrent Learning (JRL) model for exploring attribute context and correlation in order to improve attribute recognition given small sized training data with poor quality images. The JRL model learns jointly pedestrian attribute correlations in a pedestrian image and in particular their sequential ordering dependencies (latent high-order correlation) in an end-to-end encoder/decoder recurrent network. We demonstrate the performance advantage and robustness of the JRL model over a wide range of state-of-the-art deep models for pedestrian attribute recognition, multi-label image classification, and multi-person image annotation on two largest pedestrian attribute benchmarks PETA and RAP.

This paper proposes a network architecture to perform variable length semantic video generation using captions. We adopt a new perspective towards video generation where we allow the captions to be combined with the long-term and short-term dependencies between video frames and thus generate a video in an incremental manner. Our experiments demonstrate our network architecture's ability to distinguish between objects, actions and interactions in a video and combine them to generate videos for unseen captions. The network also exhibits the capability to perform spatio-temporal style transfer when asked to generate videos for a sequence of captions. We also show that the network's ability to learn a latent representation allows it generate videos in an unsupervised manner and perform other tasks such as action recognition.

Current methods for video description are based on encoder-decoder sentence generation using recurrent neural networks (RNNs). Recent work has demonstrated the advantages of integrating temporal attention mechanisms into these models, in which the decoder network predicts each word in the description by selectively giving more weight to encoded features from specific time frames. Such methods typically use two different types of features: image features (from an object classification model), and motion features (from an action recognition model), combined by naive concatenation in the model input. Because different feature modalities may carry task-relevant information at different times, fusing them by naive concatenation may limit the model's ability to dynamically determine the relevance of each type of feature to different parts of the description. In this paper, we incorporate audio features in addition to the image and motion features. To fuse these three modalities, we introduce a multimodal attention model that can selectively utilize features from different modalities for each word in the output description. Combining our new multimodal attention model with standard temporal attention outperforms state-of-the-art methods on two standard datasets: YouTube2Text and MSR-VTT.

In this paper, we propose an attention-aware deep reinforcement learning (ADRL) method for video face recognition, which aims to discard the misleading and confounding frames and find the focuses of attention in face videos for person recognition. We formulate the process of finding the attentions of videos as a Markov decision process and train the attention model through a deep reinforcement learning framework without using extra labels. Unlike existing attention models, our method takes information from both the image space and the feature space as the input to make better use of face information that is discarded in the feature learning process. Besides, our approach is attention-aware, which seeks different attentions of videos for the verification of different pairs of videos. Our approach achieves very competitive video face recognition performance on three widely used video face datasets.

We propose "associative domain adaptation", a novel technique for end-to-end domain adaptation with neural networks, the task of inferring class labels for an unlabeled target domain based on the statistical properties of a labeled source domain. Our training scheme follows the paradigm that in order to effectively derive class labels for the target domain, a network should produce statistically domain invariant embeddings, while minimizing the classification error on the labeled source domain. We accomplish this by reinforcing "associations" between source and target data directly in embedding space. Our method can easily be added to any existing classification network with no structural and almost no computational overhead. We demonstrate the effectiveness of our approach on various benchmarks and achieve state-of-the-art results across the board with a generic convolutional neural network architecture not specifically tuned to the respective tasks. Finally, we show that the proposed association loss produces embeddings that are more effective for domain adaptation compared to methods employing maximum mean discrepancy as a similarity measure in embedding space.

We propose "Areas of Attention", a novel attention-based model for automatic image captioning. Our approach models the dependencies between image regions, caption words, and the state of an RNN language model, using three pairwise interactions. In contrast to previous attention-based approaches that associate image regions to the RNN state, our method allows a direct association between caption words and image regions. During training these associations are inferred from image-level captions, akin to weakly-supervised object detector training. These associations help to improve captioning by localizing the corresponding regions during testing. We also propose and compare different ways of generating attention areas: CNN activation grids, object proposals, and spatial transformers nets applied in a convolutional fashion. Spatial transformers give the best results, since they allow for image specific attention areas, and can be trained jointly with the rest of the network. Our attention mechanism and spatial transformer attention areas together yield state-of-the-art results on the MSCOCO dataset.

Gatys et al. recently introduced a neural algorithm that renders a content image in the style of another image, achieving so-called style transfer. However, their framework requires a slow iterative optimization process, which limits its practical application. Fast approximations with feed-forward neural networks have been proposed to speed up neural style transfer. Unfortunately, the speed improvement comes at a cost: the network is usually tied to a fixed set of styles and cannot adapt to arbitrary new styles. In this paper, we present a simple yet effective approach that for the first time enables arbitrary style transfer in real-time. At the heart of our method is a novel adaptive instance normalization (AdaIN) layer that aligns the mean and variance of the content features with those of the style features. Our method achieves speed comparable to the fastest existing approach, without the restriction to a pre-defined set of styles. In addition, our approach allows flexible user controls such as content-style trade-off, style interpolation, color & spatial controls, all using a single feed-forward neural network.

Subspace learning is one of the most foundational tasks in computer vision with applications ranging from dimensionality reduction to data denoising. As geometric objects, subspaces have also been successfully used for efficiently representing certain types of invariant data. However, methods for subspace learning from subspace-valued data have been notably absent due to incompatibilities with standard problem formulations. To fill this void, we introduce Approximate Grassmannian Intersections (AGI), a novel geometric interpretation of subspace learning posed as finding the approximate intersection of constraint sets on a Grassmann manifold. Our approach can naturally be applied to input subspaces of varying dimension while reducing to standard subspace learning in the case of vector-valued data. Despite the nonconvexity of our problem, its globally-optimal solution can be found using a singular value decomposition. Furthermore, we also propose an efficient, general optimization approach that can incorporate additional constraints to encourage properties such as robustness. Alongside standard subspace applications, AGI also enables the novel task of transfer learning via subspace completion. We evaluate our approach on a variety of applications, demonstrating improved invariance and generalization over vector-valued alternatives.

Anticipating human intention by observing one's actions has many applications. For instance, picking up a cellphone, then a charger (actions) implies that one wants to charge the cellphone (intention). By anticipating the intention, an intelligent system can guide the user to the closest power outlet. We propose an on-wrist motion triggered sensing system for anticipating daily intentions, where the on-wrist sensors help us to persistently observe one's actions. The core of the system is a novel Recurrent Neural Network (RNN) and Policy Network (PN), where the RNN encodes visual and motion observation to anticipate intention, and the PN parsimoniously triggers the process of visual observation to reduce computation requirement. We jointly trained the whole network using policy gradient and cross-entropy loss. To evaluate, we collect the first daily "intention" dataset consisting of 2379 videos with 34 intentions and 164 unique action sequences. Our method achieves 92.68%, 90.85%, 97.56% accuracy on three users while processing only 29% of the visual observation on average.

To compress large datasets of high-dimensional descriptors, modern quantization schemes learn multiple codebooks and then represent individual descriptors as combinations of codewords. Once the codebooks are learned, these schemes encode descriptors independently. In contrast to that, we present a new coding scheme that arranges dataset descriptors into a set of arborescence graphs, and then encodes non-root descriptors by quantizing their displacements with respect to their parent nodes. By optimizing the structure of arborescences, our coding scheme can decrease the quantization error considerably, while incurring only minimal overhead on the memory footprint and the speed of nearest neighbor search in the compressed dataset compared to the independent quantization. The advantage of the proposed scheme is demonstrated in a series of experiments with datasets of SIFT and deep descriptors.

We propose the Anchored Regression Network (ARN), a nonlinear regression network which can be seamlessly integrated into various networks or can be used stand-alone when the features have already been fixed. Our ARN is a smoothed relaxation of a piecewise linear regressor through the combination of multiple linear regressors over soft assignments to anchor points. When the anchor points are fixed the optimal ARN regressors can be obtained with a closed form global solution, otherwise ARN admits end-to-end learning with standard gradient based methods. We demonstrate the power of the ARN by applying it to two very diverse and challenging tasks: age prediction from face images and image super-resolution. In both cases, ARNs yield strong results.

The alterations of brain structures and functions have been considered closely correlated to the change of cognitive performance due to neurodegenerative diseases such as Alzheimer's disease. In this paper, we introduce a variational framework to compute the optimal transformation (OT) in 3D space and propose a univariate neuroimaging index based on OT to measure such alterations. We compute the OT from each image to a template and measure the Wasserstein distance between them. By comparing the distances from all the images to the common template, we obtain a concise and informative index for each image. Our framework makes use of the Newton's method, which reduces the computational cost and enables itself to be applicable to large-scale datasets. The proposed work is a generic approach and thus may be applicable to various volumetric brain images, including structural magnetic resonance (sMR) and fluorodeoxyglucose positron emission tomography (FDG-PET) images. In the classification between Alzheimer's disease patients and healthy controls, our method achieves an accuracy of 82.30% on the Alzheimer's Disease Neuroimaging Initiative (ADNI) baseline sMRI dataset and outperforms several other indices. On FDG-PET dataset, we boost the accuracy to 88.37% by leveraging pairwise Wasserstein distances. In a longitudinal study, we obtain a 5% significance with p-value = 0.0000113 in a t-test on FDG-PET. The results demonstrate a great potential of the proposed index for neuroimage analysis and the precision medicine research.

Language models based on recurrent neural networks have dominated recent image caption generation tasks. In this paper, we introduce a Language CNN model which is suitable for statistical language modeling tasks and shows competitive performance in image captioning. In contrast to previous models which predict next word based on one previous word and hidden state, our language CNN is fed with all the previous words and can model the long-range dependencies in history words, which are critical for image captioning. The effectiveness of our approach is validated on two datasets: Flickr30K and MS COCO. Our extensive experimental results show that our method outperforms the vanilla recurrent neural network based language models and is competitive with the state-of-the-art methods.

In visual question answering (VQA), an algorithm must answer text-based questions about images. While multiple datasets for VQA have been created since late 2014, they all have flaws in both their content and the way algorithms are evaluated on them. As a result, evaluation scores are inflated and predominantly determined by answering easier questions, making it difficult to compare different methods. In this paper, we analyze existing VQA algorithms using a new dataset called the Task Driven Image Understanding Challenge (TDIUC), which has over 1.6 million questions organized into 12 different categories. We also introduce questions that are meaningless for a given image to force a VQA system to reason about image content. We propose new evaluation schemes that compensate for over-represented question-types and make it easier to study the strengths and weaknesses of algorithms. We analyze the performance of both baseline and state-of-the-art VQA models, including multi-modal compact bilinear pooling (MCB), neural module networks, and recurrent answering units. Our experiments establish how attention helps certain categories more than others, determine which models work better than others, and explain how simple models (e.g. MLP) can surpass more complex models (MCB) by simply learning to answer large, easy question categories.

Fully convolutional neural networks (FCNs) have shown outstanding performance in many dense labeling problems. One key pillar of these successes is mining relevant information from features in convolutional layers. However, how to better aggregate multi-level convolutional feature maps for salient object detection is underexplored. In this work, we present Amulet, a generic aggregating multi-level convolutional feature framework for salient object detection. Our framework first integrates multi-level feature maps into multiple resolutions, which simultaneously incorporate coarse semantics and fine details. Then it adaptively learns to combine these feature maps at each resolution and predict saliency maps with the combined features. Finally, the predicted results are efficiently fused to generate the final saliency map. In addition, to achieve accurate boundary inference and semantic enhancement, edge-aware feature maps in low-level layers and the predicted results of low resolution features are recursively embedded into the learning framework. By aggregating multi-level convolutional features in this efficient and flexible manner, the proposed saliency model provides accurate salient object labeling. Comprehensive experiments demonstrate that our method performs favorably against state-of-the-art approaches in terms of near all compared evaluation metrics.

This paper presents a method to assess a basketball player's performance from his/her first-person video. A key challenge lies in the fact that the evaluation metric is highly subjective and specific to a particular evaluator. We leverage the first-person camera to address this challenge. The spatiotemporal visual semantics provided by a first-person view allows us to reason about the camera wearer's actions while he/she is participating in an unscripted basketball game. Our method takes a player's first-person video and provides a player's performance measure that is specific to an evaluator's preference. To achieve this goal, we first use a convolutional LSTM network to detect atomic basketball events from first-person videos. Our network's ability to zoom-in to the salient regions addresses the issue of a severe camera wearer's head movement in first-person videos. The detected atomic events are then passed through the Gaussian mixtures to construct a highly non-linear visual spatiotemporal basketball assessment feature. Finally, we use this feature to learn a basketball assessment model from pairs of labeled first-person basketball videos, for which a basketball expert indicates, which of the two players is better. We demonstrate that despite not knowing the basketball evaluator's criterion, our model learns to accurately assess the players in real-world games. Furthermore, our model can also discover basketball events that contribute positively and negatively to a player's performance.

An important goal of computer vision is to build systems that learn visual representations over time that can be applied to many tasks. In this paper, we investigate a vision-language embedding as a core representation and show that it leads to better cross-task transfer than standard multi-task learning. In particular, the task of visual recognition is aligned to the task of visual question answering by forcing each to use the same word-region embeddings. We show this leads to greater inductive transfer from recognition to VQA than standard multitask learning. Visual recognition also improves, especially for categories that have relatively few recognition training labels but appear often in the VQA setting. Thus, our paper takes a small step towards creating more general vision systems by showing the benefit of interpretable, flexible, and trainable core representations.

It is said that a picture is worth a thousand words. Thus, there are various ways to describe an image, especially in aesthetic quality analysis. Although aesthetic quality assessment has generated a great deal of interest in the last decade, most studies focus on providing a quality rating of good or bad for an image. In this work, we extend the task to produce captions related to photo aesthetics and/or photography skills. To the best of our knowledge, this is the first study that deals with aesthetics captioning instead of AQ scoring. In contrast to common image captioning tasks that depict the objects or their relations in a picture, our approach can select a particular aesthetics aspect and generate captions with respect to the aspect chosen. Meanwhile, the proposed aspect-fusion method further uses an attention mechanism to generate more abundant aesthetics captions. We also introduce a new dataset for aesthetics captioning called the Photo Critique Captioning Dataset (PCCD), which contains pair-wise image-comment data from professional photographers. The results of experiments on PCCD demonstrate that our approaches outperform existing methods for generating aesthetic-oriented captions for images.