ICCV 2017 Papers

We present a method for Modeling Urban Scenes from Pointclouds (MUSP). In contrast to existing approaches, MUSP is robust, scalable and provides a more complete description by not making a Manhattan-World assumption and modeling both buildings (with polyhedra) as well as the non-planar ground (using NURBS). First, we segment the scene into consistent patches using a divide-and-conquer based algorithm within a nonparametric Bayesian framework (stick-breaking construction). These patches often correspond to meaningful structures, such as the ground, facades, roofs and roof superstructures. We use polygon sweeping to fit predefined templates for buildings, and for the ground, a NURBS surface is fit and uniformly tessellated. Finally, we apply boolean operations to the polygons for buildings, buildings parts and the tesselated ground to clip unnecessary geometry (e.g., facades protrusions below the non-planar ground), leading to the final model. The explicit Bayesian formulation of scene segmentation makes our approach suitable for challenging datasets with varying amounts of noise, outliers, and point density. We demonstrate the robustness of MUSP on 3D pointclouds from image matching as well as LiDAR.

In this work we propose a novel model-based deep convolutional autoencoder that addresses the highly challenging problem of reconstructing a 3D human face from a single in-the-wild color image. To this end, we combine a convolutional encoder network with an expert-designed generative model that serves as decoder. The core innovation is the differentiable parametric decoder that encapsulates image formation analytically based on a generative model. Our decoder takes as input a code vector with exactly defined semantic meaning that encodes detailed face pose, shape, expression, skin reflectance and scene illumination. Due to this new way of combining CNN-based with model-based face reconstruction, the CNN-based encoder learns to extract semantically meaningful parameters from a single monocular input image. For the first time, a CNN encoder and an expert-designed generative model can be trained end-to-end in an unsupervised manner, which renders training on very large (unlabeled) real world data feasible. The obtained reconstructions compare favorably to current state-of-the-art approaches in terms of quality and richness of representation.

Although several powerful joint filters for cross-modal image pairs have been proposed, the existing joint filters generate severe artifacts when there are misalignments between a target and a guidance images. Our goal is to generate an artifact-free output image even from the misaligned target and guidance images. We propose a novel misalignment-robust joint filter based on weight-volume-based image composition and joint-filter cost volume. Our proposed method first generates a set of translated guidances. Next, the joint-filter cost volume and a set of filtered images are computed from the target image and the set of the translated guidances. Then, a weight volume is obtained from the joint-filter cost volume while considering a spatial smoothness and a label-sparseness. The final output image is composed by fusing the set of the filtered images with the weight volume for the filtered images. The key is to generate the final output image directly from the set of the filtered images by weighted averaging using the weight volume that is obtained from the joint-filter cost volume. The proposed framework is widely applicable and can involve any kind of joint filter. Experimental results show that the proposed method is effective for various applications including image denosing, image up-sampling, haze removal and depth map interpolation.

Optical flow estimation is one of the most studied problems in computer vision, yet recent benchmark datasets continue to reveal problem areas of today's approaches. Occlusions have remained one of the key challenges. In this paper, we propose a symmetric optical flow method to address the well-known chicken-and-egg relation between optical flow and occlusions. In contrast to many state-of-the-art methods that consider occlusions as outliers, possibly filtered out during post-processing, we highlight the importance of joint occlusion reasoning in the optimization and show how to utilize occlusion as an important cue for estimating optical flow. The key feature of our model is to fully exploit the symmetry properties that characterize optical flow and occlusions in the two consecutive images. Specifically through utilizing forward-backward consistency and occlusion-disocclusion symmetry in the energy, our model jointly estimates optical flow in both forward and backward direction, as well as consistent occlusion maps in both views. We demonstrate significant performance benefits on standard benchmarks, especially from the occlusion-disocclusion symmetry. On the challenging KITTI dataset we report the most accurate two-frame results to date.

Recently, very deep convolutional neural networks (CNNs) have been attracting considerable attention in image restoration. However, as the depth grows, the long-term dependency problem is rarely realized for these very deep models, which results in the prior states/layers having little influence on the subsequent ones. Motivated by the fact that human thoughts have persistency, we propose a very deep persistent memory network (MemNet) that introduces a memory block, consisting of a recursive unit and a gate unit, to explicitly mine persistent memory through an adaptive learning process. The recursive unit learns multi-level representations of the current state under different receptive fields. The representations and the outputs from the previous memory blocks are concatenated and sent to the gate unit, which adaptively controls how much of the previous states should be reserved, and decides how much of the current state should be stored. We apply MemNet to three image restoration tasks, i.e., image denosing, super-resolution and JPEG deblocking. Comprehensive experiments demonstrate the necessity of the MemNet and its unanimous superiority on all three tasks over the state of the arts. Code is available at https://github.com/tyshiwo/MemNet.

The ability to edit materials of objects in images is desirable by many content creators. However, this is an extremely challenging task as it requires to disentangle intrinsic physical properties of an image. We propose an end-to-end network architecture that replicates the forward image formation process to accomplish this task. Specifically, given a single image, the network first predicts intrinsic properties, i.e. shape, illumination, and material, which are then provided to a rendering layer. This layer performs in-network image synthesis, thereby enabling the network to understand the physics behind the image formation process. The proposed rendering layer is fully differentiable, supports both diffuse and specular materials, and thus can be applicable in a variety of problem settings. We demonstrate a rich set of visually plausible material editing examples and provide an extensive comparative study.

We present a conceptually simple, flexible, and general framework for object instance segmentation. Our approach efficiently detects objects in an image while simultaneously generating a high-quality segmentation mask for each instance. The method, called Mask R-CNN, extends Faster R-CNN by adding a branch for predicting an object mask in parallel with the existing branch for bounding box recognition. Mask R-CNN is simple to train and adds only a small overhead to Faster R-CNN, running at 5 fps. Moreover, Mask R-CNN is easy to generalize to other tasks, e.g., allowing us to estimate human poses in the same framework. We show top results in all three tracks of the COCO suite of challenges, including instance segmentation, bounding-box object detection, and person keypoint detection. Without tricks, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners. We hope our simple and effective approach will serve as a solid baseline and help ease future research in instance-level recognition. Code will be made available.

We present a framework to analyze various aspects of models for video question answering (VideoQA) using customizable synthetic datasets, which are constructed automatically from gameplay videos. Our work is motivated by the fact that existing models are often tested only on datasets that require excessively high-level reasoning or mostly contain instances accessible through single frame inferences. Hence, it is difficult to measure capacity and flexibility of trained models, and existing techniques often rely on ad-hoc implementations of deep neural networks without clear insight into datasets and models. We are particularly interested in understanding temporal relationships between video events to solve VideoQA problems; this is because reasoning temporal dependency is one of the most distinct components in videos from images. To address this objective, we automatically generate a customized synthetic VideoQA dataset using Super Mario Bros. gameplay videos so that it contains events with different levels of reasoning complexity. Using the dataset, we show that properly constructed datasets with events in various complexity levels are critical to learn effective models and improve overall performance.

In this paper we present new techniques for constructing compact and robust minimal solvers for absolute pose estimation. We focus on the P4Pfr problem, but the methods we propose are applicable to a more general setting. Previous approaches to P4Pfr suffer from artificial degeneracies which come from their formulation and not the geometry of the original problem. In this paper we show how to avoid these false degeneracies to create more robust solvers. Combined with recently published techniques for Grobner basis solvers we are also able to construct solvers which are significantly smaller. We evaluate our solvers on both real and synthetic data, and show improved performance compared to competing solvers. Finally we show that our techniques can be directly applied to the P3.5Pf problem to get a non-degenerate solver, which is competitive with the current state-of-the-art.

Virtual face beautification (or markup) becomes common operations in camera or image processing Apps, which is actually deceiving. In this paper, we propose the task of restoring a portrait image from this process. As the first attempt along this line, we assume unknown global operations on human faces and aim to tackle the two issues of skin smoothing and skin color change. These two tasks, intriguingly, impose very different difficulties to estimate subtle details and major color variation. We propose a Component Regression Network (CRN) and address the limitation of using Euclidean loss in blind reversion. CRN maps the edited portrait images back to the original ones without knowing beautification operation details. Our experiments demonstrate effectiveness of the system for this novel task.

Bilinear models provide an appealing framework for mixing and merging information in Visual Question Answering (VQA) tasks. They help to learn high level associations between question meaning and visual concepts in the image, but they suffer from huge dimensionality issues. We introduce MUTAN, a multimodal tensor-based Tucker decomposition to efficiently parametrize bilinear interactions between visual and textual representations. Additionally to the Tucker framework, we design a low-rank matrix-based decomposition to explicitly constrain the interaction rank. With MUTAN, we control the complexity of the merging scheme while keeping nice interpretable fusion relations. We show how the Tucker decomposition framework generalizes some of the latest VQA architectures, providing state-of-the-art results.

Learning-based hashing methods are widely used for nearest neighbor retrieval, and recently, online hashing methods have demonstrated good performance-complexity trade-offs by learning hash functions from streaming data. In this paper, we first address a key challenge for online hashing: the binary codes for indexed data must be recomputed to keep pace with updates to the hash functions. We propose an efficient quality measure for hash functions, based on an information-theoretic quantity, mutual information, and use it successfully as a criterion to eliminate unnecessary hash table updates. Next, we also show how to optimize the mutual information objective using stochastic gradient descent. We thus develop a novel hashing method, MIHash, that can be used in both online and batch settings. Experiments on image retrieval benchmarks (including a 2.5M image dataset) confirm the effectiveness of our formulation, both in reducing hash table recomputations and in learning high-quality hash functions.

Low-shot visual learning--the ability to recognize novel object categories from very few examples--is a hallmark of human visual intelligence. Existing machine learning approaches fail to generalize in the same way. To make progress on this foundational problem, we present a low- shot learning benchmark on complex images that mimics challenges faced by recognition systems in the wild. We then propose (1) representation regularization techniques, and (2) techniques to hallucinate additional training examples for data-starved classes. Together, our methods improve the effectiveness of convolutional networks in low-shot learning, improving the one-shot accuracy on novel classes by 2.3x on the challenging ImageNet dataset.

Low rank tensor completion, which solves a linear inverse problem with the principle of parsimony, is a powerful technique used in many application domains in computer vision and pattern recognition. As a surrogate function of the matrix rank that is non-convex and discontinuous, the nuclear norm is often used instead to derive efficient algorithms for recovering missing information in matrices and higher order tensors. However, the nuclear norm is a loose approximation of the matrix rank, and what is more, the tensor nuclear norm is not guaranteed to be the tightest convex envelope of a multilinear rank. Alternative algorithms either require specifying/tuning several parameters (e.g., the tensor rank), and/or have a performance far from reaching the theoretical limit where the number of observed elements equals the degree of freedom in the unknown low-rank tensor. In this paper, we propose a pseudo-Bayesian approach, where a Bayesian-inspired cost function is adjusted using appropriate approximations that lead to desirable attributes including concavity and symmetry. Although deviating from the original Bayesian model, the resulting non-convex cost function is proved to have the ability to recover the true tensor with a low multilinear rank. A computational efficient algorithm is derived to solve the resulting non-convex optimization problem. We demonstrate the superior performance of the proposed algorithm in comparison with state-of-the-art alternatives by conducting extensive experiments on both synthetic data and several visual data recovery tasks.

We present a robust algorithm for personalizing a sphere-mesh tracking model to a user from a collection of depth measurements. Our core contribution is to demonstrate how simple geometric reasoning can be exploited to build a shape-space, and how its performance is comparable to shape-spaces constructed from datasets of carefully calibrated models. We achieve this goal by first re-parameterizing the geometry of the tracking template, and introducing a multi-stage calibration optimization. Our novel parameterization decouples the degrees of freedom for pose and shape, resulting in improved convergence properties. Our analytically differentiable multi-stage calibration pipeline optimizes for the model in the natural low-dimensional space of local anisotropic scalings, leading to an effective solution that can be easily embedded in other tracking/calibration algorithms. Compared to existing sphere-mesh calibration algorithms, quantitative experiments assess our algorithm possesses a larger convergence basin, and our personalized models allows to perform motion tracking with superior accuracy. Code and data are available at http://github.com/edoRemelli/hadjust

Numerous computer vision problems such as stereo depth estimation, object-class segmentation and foreground/background segmentation can be formulated as per-pixel image labeling tasks. Given one or many images as input, the desired output of these methods is usually a spatially smooth assignment of labels. The large amount of such computer vision problems has lead to significant research efforts, with the state of art moving from CRF-based approaches to deep CNNs and more recently, hybrids of the two. Although these approaches have significantly advanced the state of the art, the vast majority has solely focused on improving quantitative results and are not designed for low-compute scenarios. In this paper, we present a new general framework for a variety of computer vision labeling tasks, called HashMatch. Our approach is designed to be both fully parallel, i.e. each pixel is independently processed, and low-compute, with a model complexity an order of magnitude less than existing CNN and CRF-based approaches. We evaluate HashMatch extensively on several problems such as disparity estimation, image retrieval, feature approximation and background subtraction, for which HashMatch achieves high computational efficiency while producing high quality results.

Recently, CNN-based models have achieved remarkable success in image-based salient object detection (SOD). In these models, a key issue is to find a proper network architecture that best fits for the task of SOD. Toward this end, this paper proposes two-stream fixation-semantic CNNs, whose architecture is inspired by the fact that salient objects in complex images can be unambiguously annotated by selecting the pre-segmented semantic objects that receive the highest fixation density in eye-tracking experiments. In the two-stream CNNs, a fixation stream is pre-trained on eye-tracking data whose architecture well fits for the task of fixation prediction, and a semantic stream is pre-trained on images with semantic tags that has a proper architecture for semantic perception. By fusing these two streams into an inception-segmentation module and jointly fine-tuning them on images with manually annotated salient objects, the proposed networks show impressive performance in segmenting salient objects. Experimental results show that our approach outperforms 10 state-of-the-art models (5 deep, 5 non-deep) on 4 datasets.

We consider the question: what can be learnt by looking at and listening to a large number of unlabelled videos? There is a valuable, but so far untapped, source of information contained in the video itself -- the correspondence between the visual and the audio streams, and we introduce a novel "Audio-Visual Correspondence" learning task that makes use of this. Training visual and audio networks from scratch, without any additional supervision other than the raw unconstrained videos themselves, is shown to successfully solve this task, and, more interestingly, result in good visual and audio representations. These features set the new state-of-the-art on two sound classification benchmarks, and perform on par with the state-of-the-art self-supervised approaches on ImageNet classification. We also demonstrate that the network is able to localize objects in both modalities, as well as perform fine-grained recognition tasks.

One-shot pose estimation for tasks such as body joint localization, camera pose estimation, and object tracking are generally noisy, and temporal filters have been extensively used for regularization. One of the most widely-used methods is the Kalman filter, which is both extremely simple and general. However, Kalman filters require a motion model and measurement model to be specified a priori, which burdens the modeler and simultaneously demands that we use explicit models that are often only crude approximations of reality. For example, in the pose-estimation tasks mentioned above, it is common to use motion models that assume constant velocity or constant acceleration, and we believe that these simplified representations are severely inhibitive. In this work, we propose to instead learn rich, dynamic representations of the motion and noise models. In particular, we propose learning these models from data using long short-term memory, which allows representations that depend on all previous observations and all previous states. We evaluate our method using three of the most popular pose estimation tasks in computer vision, and in all cases we obtain state-of-the-art performance.