Computer Vision – ECCV 2018

This paper addresses the semantic instance segmentation task in the open-set conditions, where input images can contain known and unknown object classes. The training process of existing semantic instance segmentation methods requires annotation masks for all object instances, which is expensive to acquire or even infeasible in some realistic scenarios, where the number of categories may increase boundlessly. In this paper, we present a novel open-set semantic instance segmentation approach capable of segmenting all known and unknown object classes in images, based on the output of an object detector trained on known object classes. We formulate the problem using a Bayesian framework, where the posterior distribution is approximated with a simulated annealing optimization equipped with an efficient image partition sampler. We show empirically that our method is competitive with state-of-the-art supervised methods on known classes, but also performs well on unknown classes when compared with unsupervised methods.

We propose a benchmark for 6D pose estimation of a rigid object from a single RGB-D input image. The training data consists of a texture-mapped 3D object model or images of the object in known 6D poses. The benchmark comprises of: (i) eight datasets in a unified format that cover different practical scenarios, including two new datasets focusing on varying lighting conditions, (ii) an evaluation methodology with a pose-error function that deals with pose ambiguities, (iii) a comprehensive evaluation of 15 diverse recent methods that captures the status quo of the field, and (iv) an online evaluation system that is open for continuous submission of new results. The evaluation shows that methods based on point-pair features currently perform best, outperforming template matching methods, learning-based methods and methods based on 3D local features. The project website is available at bop.felk.cvut.cz .

We present a framework to reconstruct three-dimensional vehicle trajectories using monocular video data. We track two-dimensional vehicle shapes on pixel level exploiting instance-aware semantic segmentation techniques and optical flow cues. We apply Structure from Motion techniques to vehicle and background images to determine for each frame camera poses relative to vehicle instances and background structures. By combining vehicle and background camera pose information, we restrict the vehicle trajectory to a one-parameter family of possible solutions. We compute a ground representation by fusing background structures and corresponding semantic segmentations. We propose a novel method to determine vehicle trajectories consistent to image observations and reconstructed environment structures as well as a criterion to identify frames suitable for scale ratio estimation. We show qualitative results using drone imagery as well as driving sequences from the Cityscape dataset. Due to the lack of suitable benchmark datasets we present a new dataset to evaluate the quality of reconstructed three-dimensional vehicle trajectories. The video sequences show vehicles in urban areas and are rendered using the path-tracing render engine Cycles. In contrast to previous work, we perform a quantitative evaluation of the presented approach. Our algorithm achieves an average reconstruction-to-ground-truth-trajectory distance of 0.31 m using this dataset. The dataset including evaluation scripts will be publicly available on our website (Project page: http://s.fhg.de/trajectory ).

In human-object interactions (HOI) recognition, conventional methods consider the human body as a whole and pay a uniform attention to the entire body region. They ignore the fact that normally, human interacts with an object by using some parts of the body. In this paper, we argue that different body parts should be paid with different attention in HOI recognition, and the correlations between different body parts should be further considered. This is because our body parts always work collaboratively. We propose a new pairwise body-part attention model which can learn to focus on crucial parts, and their correlations for HOI recognition. A novel attention based feature selection method and a feature representation scheme that can capture pairwise correlations between body parts are introduced in the model. Our proposed approach achieved $$\mathbf {10}\%$$ 10 % relative improvement (36.1 mAP $$\rightarrow $$ → 39.9 mAP) over the state-of-the-art results in HOI recognition on the HICO dataset. We will make our model and source codes publicly available.

In this work, we address the problem of 3D human pose estimation from a sequence of 2D human poses. Although the recent success of deep networks has led many state-of-the-art methods for 3D pose estimation to train deep networks end-to-end to predict from images directly, the top-performing approaches have shown the effectiveness of dividing the task of 3D pose estimation into two steps: using a state-of-the-art 2D pose estimator to estimate the 2D pose from images and then mapping them into 3D space. They also showed that a low-dimensional representation like 2D locations of a set of joints can be discriminative enough to estimate 3D pose with high accuracy. However, estimation of 3D pose for individual frames leads to temporally incoherent estimates due to independent error in each frame causing jitter. Therefore, in this work we utilize the temporal information across a sequence of 2D joint locations to estimate a sequence of 3D poses. We designed a sequence-to-sequence network composed of layer-normalized LSTM units with shortcut connections connecting the input to the output on the decoder side and imposed temporal smoothness constraint during training. We found that the knowledge of temporal consistency improves the best reported result on Human3.6M dataset by approximately $$12.2\%$$ 12.2 % and helps our network to recover temporally consistent 3D poses over a sequence of images even when the 2D pose detector fails.

In this paper, we study the problem of recovering 3D planar surfaces from a single image of man-made environment. We show that it is possible to directly train a deep neural network to achieve this goal. A novel plane structure-induced loss is proposed to train the network to simultaneously predict a plane segmentation map and the parameters of the 3D planes. Further, to avoid the tedious manual labeling process, we show how to leverage existing large-scale RGB-D dataset to train our network without explicit 3D plane annotations, and how to take advantage of the semantic labels come with the dataset for accurate planar and non-planar classification. Experiment results demonstrate that our method significantly outperforms existing methods, both qualitatively and quantitatively. The recovered planes could potentially benefit many important visual tasks such as vision-based navigation and human-robot interaction.

Group activity recognition plays a fundamental role in a variety of applications, e.g. sports video analysis and intelligent surveillance. How to model the spatio-temporal contextual information in a scene still remains a crucial yet challenging issue. We propose a novel attentive semantic recurrent neural network (RNN), dubbed as stagNet, for understanding group activities in videos, based on the spatio-temporal attention and semantic graph. A semantic graph is explicitly modeled to describe the spatial context of the whole scene, which is further integrated with the temporal factor via structural-RNN. Benefiting from the ‘factor sharing’ and ‘message passing’ mechanisms, our model is capable of extracting discriminative spatio-temporal features and capturing inter-group relationships. Moreover, we adopt a spatio-temporal attention model to attend to key persons/frames for improved performance. Two widely-used datasets are employed for performance evaluation, and the extensive results demonstrate the superiority of our method.

Zero-shot learning (ZSL) aims to recognize objects of novel classes without any training samples of specific classes, which is achieved by exploiting the semantic information and auxiliary datasets. Recently most ZSL approaches focus on learning visual-semantic embeddings to transfer knowledge from the auxiliary datasets to the novel classes. However, few works study whether the semantic information is discriminative or not for the recognition task. To tackle such problem, we propose a coupled dictionary learning approach to align the visual-semantic structures using the class prototypes, where the discriminative information lying in the visual space is utilized to improve the less discriminative semantic space. Then, zero-shot recognition can be performed in different spaces by the simple nearest neighbor approach using the learned class prototypes. Extensive experiments on four benchmark datasets show the effectiveness of the proposed approach.

We present a simple yet efficient approach capable of training deep neural networks on large-scale weakly-supervised web images, which are crawled raw from the Internet by using text queries, without any human annotation. We develop a principled learning strategy by leveraging curriculum learning, with the goal of handling a massive amount of noisy labels and data imbalance effectively. We design a new learning curriculum by measuring the complexity of data using its distribution density in a feature space, and rank the complexity in an unsupervised manner. This allows for an efficient implementation of curriculum learning on large-scale web images, resulting in a high-performance CNN the model, where the negative impact of noisy labels is reduced substantially. Importantly, we show by experiments that those images with highly noisy labels can surprisingly improve the generalization capability of model, by serving as a manner of regularization. Our approaches obtain state-of-the-art performance on four benchmarks: WebVision, ImageNet, Clothing-1M and Food-101. With an ensemble of multiple models, we achieved a top-5 error rate of 5.2% on the WebVision challenge [18] for 1000-category classification. This result was the top performance by a wide margin, outperforming second place by a nearly 50% relative error rate. Code and models are available at: https://github.com/MalongTech/CurriculumNet .

Consumer depth sensors are more and more popular and come to our daily lives marked by its recent integration in the latest Iphone X. However, they still suffer from heavy noises which limit their applications. Although plenty of progresses have been made to reduce the noises and boost geometric details, due to the inherent illness and the real-time requirement, the problem is still far from been solved. We propose a cascaded Depth Denoising and Refinement Network (DDRNet) to tackle this problem by leveraging the multi-frame fused geometry and the accompanying high quality color image through a joint training strategy. The rendering equation is exploited in our network in an unsupervised manner. In detail, we impose an unsupervised loss based on the light transport to extract the high-frequency geometry. Experimental results indicate that our network achieves real-time single depth enhancement on various categories of scenes. Thanks to the well decoupling of the low and high frequency information in the cascaded network, we achieve superior performance over the state-of-the-art techniques.

Recent studies on face attribute transfer have achieved great success. A lot of models are able to transfer face attributes with an input image. However, they suffer from three limitations: (1) incapability of generating image by exemplars; (2) being unable to transfer multiple face attributes simultaneously; (3) low quality of generated images, such as low-resolution or artifacts. To address these limitations, we propose a novel model which receives two images of opposite attributes as inputs. Our model can transfer exactly the same type of attributes from one image to another by exchanging certain part of their encodings. All the attributes are encoded in a disentangled manner in the latent space, which enables us to manipulate several attributes simultaneously. Besides, our model learns the residual images so as to facilitate training on higher resolution images. With the help of multi-scale discriminators for adversarial training, it can even generate high-quality images with finer details and less artifacts. We demonstrate the effectiveness of our model on overcoming the above three limitations by comparing with other methods on the CelebA face database. A pytorch implementation is available at https://github.com/Prinsphield/ELEGANT .

We propose a dynamic filtering strategy with large sampling field for ConvNets (LS-DFN), where the position-specific kernels learn from not only the identical position but also multiple sampled neighbour regions. During sampling, residual learning is introduced to ease training and an attention mechanism is applied to fuse features from different samples. Such multiple samples enlarge the kernels’ receptive fields significantly without requiring more parameters. While LS-DFN inherits the advantages of DFN [5], namely avoiding feature map blurring by positionwise kernels while keeping translation invariance, it also efficiently alleviates the overfitting issue caused by much more parameters than normal CNNs. Our model is efficient and can be trained end-to-end via standard back-propagation. We demonstrate the merits of our LS-DFN on both sparse and dense prediction tasks involving object detection, semantic segmentation and flow estimation. Our results show LS-DFN enjoys stronger recognition abilities in object detection and semantic segmentation tasks on VOC benchmark [8] and sharper responses in flow estimation on FlyingChairs dataset [6] compared to strong baselines.

Due to the emergence of Generative Adversarial Networks, video synthesis has witnessed exceptional breakthroughs. However, existing methods lack a proper representation to explicitly control the dynamics in videos. Human pose, on the other hand, can represent motion patterns intrinsically and interpretably, and impose the geometric constraints regardless of appearance. In this paper, we propose a pose guided method to synthesize human videos in a disentangled way: plausible motion prediction and coherent appearance generation. In the first stage, a Pose Sequence Generative Adversarial Network (PSGAN) learns in an adversarial manner to yield pose sequences conditioned on the class label. In the second stage, a Semantic Consistent Generative Adversarial Network (SCGAN) generates video frames from the poses while preserving coherent appearances in the input image. By enforcing semantic consistency between the generated and ground-truth poses at a high feature level, our SCGAN is robust to noisy or abnormal poses. Extensive experiments on both human action and human face datasets manifest the superiority of the proposed method over other state-of-the-arts.

Deep Neural Networks (DNNs) have been widely applied in various recognition tasks. However, recently DNNs have been shown to be vulnerable against adversarial examples, which can mislead DNNs to make arbitrary incorrect predictions. While adversarial examples are well studied in classification tasks, other learning problems may have different properties. For instance, semantic segmentation requires additional components such as dilated convolutions and multiscale processing. In this paper, we aim to characterize adversarial examples based on spatial context information in semantic segmentation. We observe that spatial consistency information can be potentially leveraged to detect adversarial examples robustly even when a strong adaptive attacker has access to the model and detection strategies. We also show that adversarial examples based on attacks considered within the paper barely transfer among models, even though transferability is common in classification. Our observations shed new light on developing adversarial attacks and defenses to better understand the vulnerabilities of DNNs.

In this paper, we propose a novel joint Task-Recursive Learning (TRL) framework for the closing-loop semantic segmentation and monocular depth estimation tasks. TRL can recursively refine the results of both tasks through serialized task-level interactions. In order to mutually-boost for each other, we encapsulate the interaction into a specific Task-Attentional Module (TAM) to adaptively enhance some counterpart patterns of both tasks. Further, to make the inference more credible, we propagate previous learning experiences on both tasks into the next network evolution by explicitly concatenating previous responses. The sequence of task-level interactions are finally evolved along a coarse-to-fine scale space such that the required details may be reconstructed progressively. Extensive experiments on NYU-Depth v2 and SUN RGB-D datasets demonstrate that our method achieves state-of-the-art results for monocular depth estimation and semantic segmentation.

In recent years, deep learning methods have been successfully applied to single-image super-resolution tasks. Despite their great performances, deep learning methods cannot be easily applied to real-world applications due to the requirement of heavy computation. In this paper, we address this issue by proposing an accurate and lightweight deep network for image super-resolution. In detail, we design an architecture that implements a cascading mechanism upon a residual network. We also present variant models of the proposed cascading residual network to further improve efficiency. Our extensive experiments show that even with much fewer parameters and operations, our models achieve performance comparable to that of state-of-the-art methods.

Modern semantic segmentation frameworks usually combine low-level and high-level features from pre-trained backbone convolutional models to boost performance. In this paper, we first point out that a simple fusion of low-level and high-level features could be less effective because of the gap in semantic levels and spatial resolution. We find that introducing semantic information into low-level features and high-resolution details into high-level features is more effective for the later fusion. Based on this observation, we propose a new framework, named ExFuse, to bridge the gap between low-level and high-level features thus significantly improve the segmentation quality by 4.0% in total. Furthermore, we evaluate our approach on the challenging PASCAL VOC 2012 segmentation benchmark and achieve 87.9% mean IoU, which outperforms the previous state-of-the-art results.

This work proposes an algorithm, called NetAdapt, that automatically adapts a pre-trained deep neural network to a mobile platform given a resource budget. While many existing algorithms simplify networks based on the number of MACs or weights, optimizing those indirect metrics may not necessarily reduce the direct metrics, such as latency and energy consumption. To solve this problem, NetAdapt incorporates direct metrics into its adaptation algorithm. These direct metrics are evaluated using empirical measurements, so that detailed knowledge of the platform and toolchain is not required. NetAdapt automatically and progressively simplifies a pre-trained network until the resource budget is met while maximizing the accuracy. Experiment results show that NetAdapt achieves better accuracy versus latency trade-offs on both mobile CPU and mobile GPU, compared with the state-of-the-art automated network simplification algorithms. For image classification on the ImageNet dataset, NetAdapt achieves up to a 1.7 $$\times $$ × speedup in measured inference latency with equal or higher accuracy on MobileNets (V1&V2).

We introduce a novel Recurrent Neural Network-based algorithm for future video feature generation and action anticipation called feature mapping RNN. Our novel RNN architecture builds upon three effective principles of machine learning, namely parameter sharing, Radial Basis Function kernels and adversarial training. Using only some of the earliest frames of a video, the feature mapping RNN is able to generate future features with a fraction of the parameters needed in traditional RNN. By feeding these future features into a simple multilayer perceptron facilitated with an RBF kernel layer, we are able to accurately predict the action in the video.In our experiments, we obtain 18% improvement on JHMDB-21 dataset, 6% on UCF101-24 and 13% improvement on UT-Interaction datasets over prior state-of-the-art for action anticipation.

We show that, in images of man-made environments, the horizon line can usually be hypothesized based on a-contrario detections of second-order grouping events. This allows constraining the extraction of the horizontal vanishing points on that line, thus reducing false detections. Experiments made on three datasets show that our method, not only achieves state-of-the-art performance w.r.t. horizon line detection on two datasets, but also yields much less spurious vanishing points than the previous top-ranked methods.

In this work, we consider the problem of robust gaze estimation in natural environments. Large camera-to-subject distances and high variations in head pose and eye gaze angles are common in such environments. This leads to two main shortfalls in state-of-the-art methods for gaze estimation: hindered ground truth gaze annotation and diminished gaze estimation accuracy as image resolution decreases with distance. We first record a novel dataset of varied gaze and head pose images in a natural environment, addressing the issue of ground truth annotation by measuring head pose using a motion capture system and eye gaze using mobile eyetracking glasses. We apply semantic image inpainting to the area covered by the glasses to bridge the gap between training and testing images by removing the obtrusiveness of the glasses. We also present a new real-time algorithm involving appearance-based deep convolutional neural networks with increased capacity to cope with the diverse images in the new dataset. Experiments with this network architecture are conducted on a number of diverse eye-gaze datasets including our own, and in cross dataset evaluations. We demonstrate state-of-the-art performance in terms of estimation accuracy in all experiments, and the architecture performs well even on lower resolution images.

In this paper, we present an end-to-end deblurring network designed specifically for a class of data. Unlike the prior supervised deep-learning works that extensively rely on large sets of paired data, which is highly demanding and challenging to obtain, we propose an unsupervised training scheme with unpaired data to achieve the same. Our model consists of a Generative Adversarial Network (GAN) that learns a strong prior on the clean image domain using adversarial loss and maps the blurred image to its clean equivalent. To improve the stability of GAN and to preserve the image correspondence, we introduce an additional CNN module that reblurs the generated GAN output to match with the blurred input. Along with these two modules, we also make use of the blurred image itself to self-guide the network to constrain the solution space of generated clean images. This self-guidance is achieved by imposing a scale-space gradient error with an additional gradient module. We train our model on different classes and observe that adding the reblur and gradient modules helps in better convergence. Extensive experiments demonstrate that our method performs favorably against the state-of-the-art supervised methods on both synthetic and real-world images even in the absence of any supervision.

With the advantage of high mobility, Unmanned Aerial Vehicles (UAVs) are used to fuel numerous important applications in computer vision, delivering more efficiency and convenience than surveillance cameras with fixed camera angle, scale and view. However, very limited UAV datasets are proposed, and they focus only on a specific task such as visual tracking or object detection in relatively constrained scenarios. Consequently, it is of great importance to develop an unconstrained UAV benchmark to boost related researches. In this paper, we construct a new UAV benchmark focusing on complex scenarios with new level challenges. Selected from 10 hours raw videos, about 80, 000 representative frames are fully annotated with bounding boxes as well as up to 14 kinds of attributes (e.g., weather condition, flying altitude, camera view, vehicle category, and occlusion) for three fundamental computer vision tasks: object detection, single object tracking, and multiple object tracking. Then, a detailed quantitative study is performed using most recent state-of-the-art algorithms for each task. Experimental results show that the current state-of-the-art methods perform relative worse on our dataset, due to the new challenges appeared in UAV based real scenes, e.g., high density, small object, and camera motion. To our knowledge, our work is the first time to explore such issues in unconstrained scenes comprehensively. The dataset and all the experimental results are available in https://sites.google.com/site/daviddo0323/ .

Spatio-temporal representations in frame sequences play an important role in the task of action recognition. Previously, a method of using optical flow as a temporal information in combination with a set of RGB images that contain spatial information has shown great performance enhancement in the action recognition tasks. However, it has an expensive computational cost and requires two-stream (RGB and optical flow) framework. In this paper, we propose MFNet (Motion Feature Network) containing motion blocks which make it possible to encode spatio-temporal information between adjacent frames in a unified network that can be trained end-to-end. The motion block can be attached to any existing CNN-based action recognition frameworks with only a small additional cost. We evaluated our network on two of the action recognition datasets (Jester and Something-Something) and achieved competitive performances for both datasets by training the networks from scratch.

Template matching is a fundamental problem in computer vision, with many applications. Existing methods use sliding window computation for choosing an image-window that best matches the template. For classic algorithms based on SSD, SAD and normalized cross-correlation, efficient algorithms have been developed allowing them to run in real-time. Current state of the art algorithms are based on nearest neighbor (NN) matching of small patches within the template to patches in the image. These algorithms yield state-of-the-art results since they can deal better with changes in appearance, viewpoint, illumination, non-rigid transformations, and occlusion. However, NN-based algorithms are relatively slow not only due to NN computation for each image patch, but also since their sliding window computation is inefficient. We therefore propose in this paper an efficient NN-based algorithm. Its accuracy is similar (in some cases slightly better) than the existing algorithms and its running time is 43–200 times faster depending on the sizes of the images and templates used. The main contribution of our method is an algorithm for incrementally computing the score of each image window based on the score computed for the previous window. This is in contrast to computing the score for each image window independently, as in previous NN-based methods. The complexity of our method is therefore O(|I|) instead of O(|I||T|), where I and T are the image and the template respectively.

The lack of realistic and open benchmarking datasets for pedestrian visual-inertial odometry has made it hard to pinpoint differences in published methods. Existing datasets either lack a full six degree-of-freedom ground-truth or are limited to small spaces with optical tracking systems. We take advantage of advances in pure inertial navigation, and develop a set of versatile and challenging real-world computer vision benchmark sets for visual-inertial odometry. For this purpose, we have built a test rig equipped with an iPhone, a Google Pixel Android phone, and a Google Tango device. We provide a wide range of raw sensor data that is accessible on almost any modern-day smartphone together with a high-quality ground-truth track. We also compare resulting visual-inertial tracks from Google Tango, ARCore, and Apple ARKit with two recent methods published in academic forums. The data sets cover both indoor and outdoor cases, with stairs, escalators, elevators, office environments, a shopping mall, and metro station.

We consider the problem of inferring a layered representation, its depth ordering and motion segmentation from video in which objects may undergo 3D non-planar motion relative to the camera. We generalize layered inference to that case and corresponding self-occlusion phenomena. We accomplish this by introducing a flattened 3D object representation, which is a compact representation of an object that contains all visible portions of the object seen in the video, including parts of an object that are self-occluded (as well as occluded) in one frame but seen in another. We formulate the inference of such flattened representations and motion segmentation, and derive an optimization scheme. We also introduce a new depth ordering scheme, which is independent of layered inference and addresses the case of self-occlusion. It requires little computation given the flattened representations. Experiments on benchmark datasets show the advantage of our method over existing layered methods, which do not model 3D motion and self-occlusion.

We present 3DMV, a novel method for 3D semantic scene segmentation of RGB-D scans in indoor environments using a joint 3D-multi-view prediction network. In contrast to existing methods that either use geometry or RGB data as input for this task, we combine both data modalities in a joint, end-to-end network architecture. Rather than simply projecting color data into a volumetric grid and operating solely in 3D – which would result in insufficient detail – we first extract feature maps from associated RGB images. These features are then mapped into the volumetric feature grid of a 3D network using a differentiable back-projection layer. Since our target is 3D scanning scenarios with possibly many frames, we use a multi-view pooling approach in order to handle a varying number of RGB input views. This learned combination of RGB and geometric features with our joint 2D-3D architecture achieves significantly better results than existing baselines. For instance, our final result on the ScanNet 3D segmentation benchmark increases from 52.8% to 75% accuracy compared to existing volumetric architectures.

Images captured by fisheye lenses violate the pinhole camera assumption and suffer from distortions. Rectification of fisheye images is therefore a crucial preprocessing step for many computer vision applications. In this paper, we propose an end-to-end multi-context collaborative deep network for removing distortions from single fisheye images. In contrast to conventional approaches, which focus on extracting hand-crafted features from input images, our method learns high-level semantics and low-level appearance features simultaneously to estimate the distortion parameters. To facilitate training, we construct a synthesized dataset that covers various scenes and distortion parameter settings. Experiments on both synthesized and real-world datasets show that the proposed model significantly outperforms current state of the art methods. Our code and synthesized dataset will be made publicly available.

This paper addresses the single-image compressive sensing (CS) and reconstruction problem. We propose a scalable Laplacian pyramid reconstructive adversarial network (LAPRAN) that enables high-fidelity, flexible and fast CS images reconstruction. LAPRAN progressively reconstructs an image following the concept of the Laplacian pyramid through multiple stages of reconstructive adversarial networks (RANs). At each pyramid level, CS measurements are fused with a contextual latent vector to generate a high-frequency image residual. Consequently, LAPRAN can produce hierarchies of reconstructed images and each with an incremental resolution and improved quality. The scalable pyramid structure of LAPRAN enables high-fidelity CS reconstruction with a flexible resolution that is adaptive to a wide range of compression ratios (CRs), which is infeasible with existing methods. Experimental results on multiple public datasets show that LAPRAN offers an average 7.47 dB and 5.98 dB PSNR, and an average 57.93 $$\%$$ % and 33.20 $$\%$$ % SSIM improvement compared to model-based and data-driven baselines, respectively.

Reconstructing 3D facial geometry from a single RGB image has recently instigated wide research interest. However, it is still an ill-posed problem and most methods rely on prior models hence undermining the accuracy of the recovered 3D faces. In this paper, we exploit the Epipolar Plane Images (EPI) obtained from light field cameras and learn CNN models that recover horizontal and vertical 3D facial curves from the respective horizontal and vertical EPIs. Our 3D face reconstruction network (FaceLFnet) comprises a densely connected architecture to learn accurate 3D facial curves from low resolution EPIs. To train the proposed FaceLFnets from scratch, we synthesize photo-realistic light field images from 3D facial scans. The curve by curve 3D face estimation approach allows the networks to learn from only 14K images of 80 identities, which still comprises over 11 Million EPIs/curves. The estimated facial curves are merged into a single pointcloud to which a surface is fitted to get the final 3D face. Our method is model-free, requires only a few training samples to learn FaceLFnet and can reconstruct 3D faces with high accuracy from single light field images under varying poses, expressions and lighting conditions. Comparison on the BU-3DFE and BU-4DFE datasets show that our method reduces reconstruction errors by over 20% compared to recent state of the art.

Generating stylized captions for an image is an emerging topic in image captioning. Given an image as input, it requires the system to generate a caption that has a specific style (e.g., humorous, romantic, positive, and negative) while describing the image content semantically accurately. In this paper, we propose a novel stylized image captioning model that effectively takes both requirements into consideration. To this end, we first devise a new variant of LSTM, named style-factual LSTM, as the building block of our model. It uses two groups of matrices to capture the factual and stylized knowledge, respectively, and automatically learns the word-level weights of the two groups based on previous context. In addition, when we train the model to capture stylized elements, we propose an adaptive learning approach based on a reference factual model, it provides factual knowledge to the model as the model learns from stylized caption labels, and can adaptively compute how much information to supply at each time step. We evaluate our model on two stylized image captioning datasets, which contain humorous/romantic captions and positive/negative captions, respectively. Experiments shows that our proposed model outperforms the state-of-the-art approaches, without using extra ground truth supervision.

Image geolocalization is the task of identifying the location depicted in a photo based only on its visual information. This task is inherently challenging since many photos have only few, possibly ambiguous cues to their geolocation. Recent work has cast this task as a classification problem by partitioning the earth into a set of discrete cells that correspond to geographic regions. The granularity of this partitioning presents a critical trade-off; using fewer but larger cells results in lower location accuracy while using more but smaller cells reduces the number of training examples per class and increases model size, making the model prone to overfitting. To tackle this issue, we propose a simple but effective algorithm, combinatorial partitioning, which generates a large number of fine-grained output classes by intersecting multiple coarse-grained partitionings of the earth. Each classifier votes for the fine-grained classes that overlap with their respective coarse-grained ones. This technique allows us to predict locations at a fine scale while maintaining sufficient training examples per class. Our algorithm achieves the state-of-the-art performance in location recognition on multiple benchmark datasets.

We introduce a fast and efficient convolutional neural network, ESPNet, for semantic segmentation of high resolution images under resource constraints. ESPNet is based on a new convolutional module, efficient spatial pyramid (ESP), which is efficient in terms of computation, memory, and power. ESPNet is 22 times faster (on a standard GPU) and 180 times smaller than the state-of-the-art semantic segmentation network PSPNet, while its category-wise accuracy is only 8% less. We evaluated ESPNet on a variety of semantic segmentation datasets including Cityscapes, PASCAL VOC, and a breast biopsy whole slide image dataset. Under the same constraints on memory and computation, ESPNet outperforms all the current efficient CNN networks such as MobileNet, ShuffleNet, and ENet on both standard metrics and our newly introduced performance metrics that measure efficiency on edge devices. Our network can process high resolution images at a rate of 112 and 9 frames per second on a standard GPU and edge device, respectively. Our code is open-source and available at https://sacmehta.github.io/ESPNet/ .

We introduce the Densely Segmented Supermarket (D2S) dataset, a novel benchmark for instance-aware semantic segmentation in an industrial domain. It contains 21 000 high-resolution images with pixel-wise labels of all object instances. The objects comprise groceries and everyday products from 60 categories. The benchmark is designed such that it resembles the real-world setting of an automatic checkout, inventory, or warehouse system. The training images only contain objects of a single class on a homogeneous background, while the validation and test sets are much more complex and diverse. To further benchmark the robustness of instance segmentation methods, the scenes are acquired with different lightings, rotations, and backgrounds. We ensure that there are no ambiguities in the labels and that every instance is labeled comprehensively. The annotations are pixel-precise and allow using crops of single instances for articial data augmentation. The dataset covers several challenges highly relevant in the field, such as a limited amount of training data and a high diversity in the test and validation sets. The evaluation of state-of-the-art object detection and instance segmentation methods on D2S reveals significant room for improvement.

We present an end-to-end solution for recognizing merchandise displayed in the shelves of a supermarket. Given images of individual products, which are taken under ideal illumination for product marketing, the challenge is to find these products automatically in the images of the shelves. Note that the images of shelves are taken using hand-held camera under store level illumination. We provide a two-layer hypotheses generation and verification model. In the first layer, the model predicts a set of candidate merchandise at a specific location of the shelf while in the second layer, the hypothesis is verified by a novel graph theoretic approach. The performance of the proposed approach on two publicly available datasets is better than the competing approaches by at least 10%.

In this work, we propose a method that combines a single hand-held camera and a set of Inertial Measurement Units (IMUs) attached at the body limbs to estimate accurate 3D poses in the wild. This poses many new challenges: the moving camera, heading drift, cluttered background, occlusions and many people visible in the video. We associate 2D pose detections in each image to the corresponding IMU-equipped persons by solving a novel graph based optimization problem that forces 3D to 2D coherency within a frame and across long range frames. Given associations, we jointly optimize the pose of a statistical body model, the camera pose and heading drift using a continuous optimization framework. We validated our method on the TotalCapture dataset, which provides video and IMU synchronized with ground truth. We obtain an accuracy of 26 mm, which makes it accurate enough to serve as a benchmark for image-based 3D pose estimation in the wild. Using our method, we recorded 3D Poses in the Wild (3DPW), a new dataset consisting of more than 51, 000 frames with accurate 3D pose in challenging sequences, including walking in the city, going up-stairs, having coffee or taking the bus. We make the reconstructed 3D poses, video, IMU and 3D models available for research purposes at http://virtualhumans.mpi-inf.mpg.de/3DPW .

We present a novel regularization approach to train neural networks that enjoys better generalization and test error than standard stochastic gradient descent. Our approach is based on the principles of cross-validation, where a validation set is used to limit the model overfitting. We formulate such principles as a bilevel optimization problem. This formulation allows us to define the optimization of a cost on the validation set subject to another optimization on the training set. The overfitting is controlled by introducing weights on each mini-batch in the training set and by choosing their values so that they minimize the error on the validation set. In practice, these weights define mini-batch learning rates in a gradient descent update equation that favor gradients with better generalization capabilities. Because of its simplicity, this approach can be integrated with other regularization methods and training schemes. We evaluate extensively our proposed algorithm on several neural network architectures and datasets, and find that it consistently improves the generalization of the model, especially when labels are noisy.

Compressive video sensing is the process of encoding multiple sub-frames into a single frame with controlled sensor exposures and reconstructing the sub-frames from the single compressed frame. It is known that spatially and temporally random exposures provide the most balanced compression in terms of signal recovery. However, sensors that achieve a fully random exposure on each pixel cannot be easily realized in practice because the circuit of the sensor becomes complicated and incompatible with the sensitivity and resolution. Therefore, it is necessary to design an exposure pattern by considering the constraints enforced by hardware. In this paper, we propose a method of jointly optimizing the exposure patterns of compressive sensing and the reconstruction framework under hardware constraints. By conducting a simulation and actual experiments, we demonstrated that the proposed framework can reconstruct multiple sub-frame images with higher quality.

In this work we introduce Deforming Autoencoders, a generative model for images that disentangles shape from appearance in an unsupervised manner. As in the deformable template paradigm, shape is represented as a deformation between a canonical coordinate system (‘template’) and an observed image, while appearance is modeled in deformation-invariant, template coordinates. We introduce novel techniques that allow this approach to be deployed in the setting of autoencoders and show that this method can be used for unsupervised group-wise image alignment. We show experiments with expression morphing in humans, hands, and digits, face manipulation, such as shape and appearance interpolation, as well as unsupervised landmark localization. We also achieve a more powerful form of unsupervised disentangling in template coordinates, that successfully decomposes face images into shading and albedo, allowing us to further manipulate face images.

We introduce a new method for interpreting computer vision models: visually perceptible, decision-boundary crossing transformations. Our goal is to answer a simple question: why did a model classify an image as being of class A instead of class B? Existing approaches to model interpretation, including saliency and explanation-by-nearest neighbor, fail to visually illustrate examples of transformations required for a specific input to alter a model’s prediction. On the other hand, algorithms for creating decision-boundary crossing transformations (e.g., adversarial examples) produce differences that are visually imperceptible and do not enable insightful explanation. To address this we introduce ExplainGAN, a generative model that produces visually perceptible decision-boundary crossing transformations. These transformations provide high-level conceptual insights which illustrate how a model makes decisions. We validate our model using both traditional quantitative interpretation metrics and introduce a new validation scheme for our approach and generative models more generally.

Hazy images are common in real scenarios and many dehazing methods have been developed to automatically remove the haze from images. Typically, the goal of image dehazing is to produce clearer images from which human vision can better identify the object and structural details present in the images. When the ground-truth haze-free image is available for a hazy image, quantitative evaluation of image dehazing is usually based on objective metrics, such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM). However, in many applications, large-scale images are collected not for visual examination by human. Instead, they are used for many high-level vision tasks, such as automatic classification, recognition and categorization. One fundamental problem here is whether various dehazing methods can produce clearer images that can help improve the performance of the high-level tasks. In this paper, we empirically study this problem in the important task of image classification by using both synthetic and real hazy image datasets. From the experimental results, we find that the existing image-dehazing methods cannot improve much the image-classification performance and sometimes even reduce the image-classification performance.

Recent work has demonstrated that it is possible to learn deep neural networks for monocular depth and ego-motion estimation from unlabelled video sequences, an interesting theoretical development with numerous advantages in applications. In this paper, we propose a number of improvements to these approaches. First, since such self-supervised approaches are based on the brightness constancy assumption, which is valid only for a subset of pixels, we propose a probabilistic learning formulation where the network predicts distributions over variables rather than specific values. As these distributions are conditioned on the observed image, the network can learn which scene and object types are likely to violate the model assumptions, resulting in more robust learning. We also propose to build on dozens of years of experience in developing handcrafted structure-from-motion (SFM) algorithms. We do so by using an off-the-shelf SFM system to generate a supervisory signal for the deep neural network. While this signal is also noisy, we show that our probabilistic formulation can learn and account for the defects of SFM, helping to integrate different sources of information and boosting the overall performance of the network.

A vexing problem in artificial intelligence is reasoning about events that occur in complex, changing visual stimuli such as in video analysis or game play. Inspired by a rich tradition of visual reasoning and memory in cognitive psychology and neuroscience, we developed an artificial, configurable visual question and answer dataset (COG) to parallel experiments in humans and animals. COG is much simpler than the general problem of video analysis, yet it addresses many of the problems relating to visual and logical reasoning and memory – problems that remain challenging for modern deep learning architectures. We additionally propose a deep learning architecture that performs competitively on other diagnostic VQA datasets (i.e. CLEVR) as well as easy settings of the COG dataset. However, several settings of COG result in datasets that are progressively more challenging to learn. After training, the network can zero-shot generalize to many new tasks. Preliminary analyses of the network architectures trained on COG demonstrate that the network accomplishes the task in a manner interpretable to humans.

We present COBLA—Constrained Optimization Based Low-rank Approximation—a systematic method of finding an optimal low-rank approximation of a trained convolutional neural network, subject to constraints in the number of multiply-accumulate (MAC) operations and the memory footprint. COBLA optimally allocates the constrained computation resources into each layer of the approximated network. The singular value decomposition of the network weight is computed, then a binary masking variable is introduced to denote whether a particular singular value and the corresponding singular vectors are used in low-rank approximation. With this formulation, the number of the MAC operations and the memory footprint are represented as linear constraints in terms of the binary masking variables. The resulted 0–1 integer programming problem is approximately solved by sequential quadratic programming. COBLA does not introduce any hyperparameter. We empirically demonstrate that COBLA outperforms prior art using the SqueezeNet and VGG-16 architecture on the ImageNet dataset.

Modern 3D human pose estimation techniques rely on deep networks, which require large amounts of training data. While weakly-supervised methods require less supervision, by utilizing 2D poses or multi-view imagery without annotations, they still need a sufficiently large set of samples with 3D annotations for learning to succeed.In this paper, we propose to overcome this problem by learning a geometry-aware body representation from multi-view images without annotations. To this end, we use an encoder-decoder that predicts an image from one viewpoint given an image from another viewpoint. Because this representation encodes 3D geometry, using it in a semi-supervised setting makes it easier to learn a mapping from it to 3D human pose. As evidenced by our experiments, our approach significantly outperforms fully-supervised methods given the same amount of labeled data, and improves over other semi-supervised methods while using as little as 1% of the labeled data.

In this paper, we propose a dual-agent deep reinforcement learning (DADRL) method for deformable face tracking, which generates bounding boxes and detects facial landmarks interactively from face videos. Most existing deformable face tracking methods learn models for these two tasks individually, and perform these two procedures subsequently during the testing phase, which ignore the intrinsic connections of these two tasks. Motivated by the fact that the performance of facial landmark detection depends heavily on the accuracy of the generated bounding boxes, we exploit the interactions of these two tasks in probabilistic manner by following a Bayesian model and propose a unified framework for simultaneous bounding box tracking and landmark detection. By formulating it as a Markov decision process, we define two agents to exploit the relationships and pass messages via an adaptive sequence of actions under a deep reinforcement learning framework to iteratively adjust the positions of the bounding boxes and facial landmarks. Our proposed DADRL achieves performance improvements over the state-of-the-art deformable face tracking methods on the most challenging category of the 300-VW dataset.

We present a method for simultaneously estimating 3D human pose and body shape from a sparse set of wide-baseline camera views. We train a symmetric convolutional autoencoder with a dual loss that enforces learning of a latent representation that encodes skeletal joint positions, and at the same time learns a deep representation of volumetric body shape. We harness the latter to up-scale input volumetric data by a factor of 4 $$\times $$ × , whilst recovering a 3D estimate of joint positions with equal or greater accuracy than the state of the art. Inference runs in real-time (25 fps) and has the potential for passive human behaviour monitoring where there is a requirement for high fidelity estimation of human body shape and pose.

Learning and predicting the pose parameters of a 3D hand model given an image, such as locations of hand joints, is challenging due to large viewpoint changes and articulations, and severe self-occlusions exhibited particularly in egocentric views. Both feature learning and prediction modeling have been investigated to tackle the problem. Though effective, most existing discriminative methods yield a single deterministic estimation of target poses. Due to their single-value mapping intrinsic, they fail to adequately handle self-occlusion problems, where occluded joints present multiple modes. In this paper, we tackle the self-occlusion issue and provide a complete description of observed poses given an input depth image by a novel method called hierarchical mixture density networks (HMDN). The proposed method leverages the state-of-the-art hand pose estimators based on Convolutional Neural Networks to facilitate feature learning, while it models the multiple modes in a two-level hierarchy to reconcile single-valued and multi-valued mapping in its output. The whole framework with a mixture of two differentiable density functions is naturally end-to-end trainable. In the experiments, HMDN produces interpretable and diverse candidate samples, and significantly outperforms the state-of-the-art methods on two benchmarks with occlusions, and performs comparably on another benchmark free of occlusions.

Recent advances in Generative Adversarial Networks (GANs) have shown impressive results for task of facial expression synthesis. The most successful architecture is StarGAN, that conditions GANs’ generation process with images of a specific domain, namely a set of images of persons sharing the same expression. While effective, this approach can only generate a discrete number of expressions, determined by the content of the dataset. To address this limitation, in this paper, we introduce a novel GAN conditioning scheme based on Action Units (AU) annotations, which describes in a continuous manifold the anatomical facial movements defining a human expression. Our approach allows controlling the magnitude of activation of each AU and combine several of them. Additionally, we propose a fully unsupervised strategy to train the model, that only requires images annotated with their activated AUs, and exploit attention mechanisms that make our network robust to changing backgrounds and lighting conditions. Extensive evaluation show that our approach goes beyond competing conditional generators both in the capability to synthesize a much wider range of expressions ruled by anatomically feasible muscle movements, as in the capacity of dealing with images in the wild.