Rapid and reliable identification of dynamic scene parts, also known as motion segmentation, is a key challenge for mobile sensors. Contemporary RGB camera-based methods rely on modeling camera and scene properties however, are often under-constrained and fall short in unknown categories. Event cameras have the potential to overcome these limitations, but corresponding methods have only been demonstrated in smaller-scale indoor environments with simplified dynamic objects. This work presents an event-based method for class-agnostic motion segmentation that can successfully be deployed across complex large-scale outdoor environments too. To this end, we introduce a novel divide-and-conquer pipeline that combines: (a) ego-motion compensated events, computed via a scene understanding module that predicts monocular depth and camera pose as auxiliary tasks, and (b) optical flow from a dedicated optical flow module. These intermediate representations are then fed into a segmentation module that predicts motion segmentation masks. A novel transformer-based temporal attention module in the segmentation module builds correlations across adjacent ‘frames’ to get temporally consistent segmentation masks. Our method sets the new state-of-the-art on the classic EV-IMO benchmark (indoors), where we achieve improvements of 2.19 moving object IoU (2.22 mIoU) and 4.52 point IoU respectively, as well as on a newly-generated motion segmentation and tracking benchmark (outdoors) based on the DSEC event dataset, termed DSEC-MOTS, where we show improvement of 12.91 moving object IoU.

Recently, video frame interpolation using a combination of frame- and event-based cameras has surpassed traditional image-based methods both in terms of performance and memory efficiency. However, current methods still suffer from (i) brittle image-level fusion of complementary interpolation results, that fails in the presence of artifacts in the fused image, (ii) potentially temporally inconsistent and inefficient motion estimation procedures, that run for every inserted frame and (iii) low contrast regions that do not trigger events, and thus cause events-only motion estimation to generate artifacts. Moreover, previous methods were only tested on datasets consisting of planar and far-away scenes, which do not capture the full complexity of the real world. In this work, we address the above problems by introducing multi-scale feature-level fusion and computing one-shot non-linear inter-frame motion-which can be efficiently sampled for image warping-from events and images. We also collect the first large-scale events and frames dataset consisting of more than 100 challenging scenes with depth variations, captured with a new experimental setup based on a beamsplitter. We show that our method improves the reconstruction quality by up to 0.2 dB in terms of PSNR and up to 15% in LPIPS score.

With the advent of deep learning, many dense prediction tasks, i.e., tasks that produce pixel-level predictions, have seen significant performance improvements. The typical approach is to learn these tasks in isolation, that is, a separate neural network is trained for each individual task. Yet, recent multi-task learning (MTL) techniques have shown promising results w.r.t. performance, computations and/or memory footprint, by jointly tackling multiple tasks through a learned shared representation. In this survey, we provide a well-rounded view on state-of-the-art deep learning approaches for MTL in computer vision, explicitly emphasizing on dense prediction tasks. Our contributions concern the following. First, we consider MTL from a network architecture point-of-view. We include an extensive overview and discuss the advantages/disadvantages of recent popular MTL models. Second, we examine various optimization methods to tackle the joint learning of multiple tasks. We summarize the qualitative elements of these works and explore their commonalities and differences. Finally, we provide an extensive experimental evaluation across a variety of dense prediction benchmarks to examine the pros and cons of the different methods, including both architectural and optimization based strategies.

We study low-rank parameterizations of weight matrices with embedded spectral properties in the Deep Learning context. The low-rank property leads to parameter efficiency and permits taking computational shortcuts when computing mappings. Spectral properties are often subject to constraints in optimization problems, leading to better models and stability of optimization. We start by looking at the compact SVD parameterization of weight matrices and identifying redundancy sources in the parameterization. We further apply the Tensor Train (TT) decomposition to the compact SVD components, and propose a non-redundant differentiable parameterization of fixed TT-rank tensor manifolds, termed the Spectral Tensor Train Parameterization (STTP). We demonstrate the effects of neural network compression in the image classification setting, and both compression and improved training stability in the generative adversarial training setting. Project website: www.obukhov.ai/sttp

We present an approach for encoding visual task relationships to improve model performance in an Unsupervised Domain Adaptation (UDA) setting. Semantic segmentation and monocular depth estimation are shown to be complementary tasks; in a multi-task learning setting, a proper encoding of their relationships can further improve performance on both tasks. Motivated by this observation, we propose a novel Cross-Task Relation Layer (CTRL), which encodes task dependencies between the semantic and depth predictions. To capture the cross-task relationships, we propose a neural network architecture that contains task-specific and cross-task refinement heads. Furthermore, we propose an Iterative Self-Learning (ISL) training scheme, which exploits semantic pseudo-labels to provide extra supervision on the target domain. We experimentally observe improvements in both tasks’ performance because the complementary information present in these tasks is better captured. Specifically, we show that: (1) our approach improves performance on all tasks when they are complementary and mutually dependent; (2) the CTRL helps to improve both semantic segmentation and depth estimation tasks performance in the challenging UDA setting; (3) the proposed ISL training scheme further improves the semantic segmentation performance. The implementation is available at https://github.com/susaha/ctrl-uda.

State-of-the-art frame interpolation methods generate intermediate frames by inferring object motions in the image from consecutive key-frames. In the absence of additional information, first-order approximations, i.e. optical flow, must be used, but this choice restricts the types of motions that can be modeled, leading to errors in highly dynamic scenarios. Event cameras are novel sensors that address this limitation by providing auxiliary visual information in the blind-time between frames. They asynchronously measure per-pixel brightness changes and do this with high temporal resolution and low latency. Event-based frame interpolation methods typically adopt a synthesis-based approach, where predicted frame residuals are directly applied to the key-frames. However, while these approaches can capture non-linear motions they suffer from ghosting and perform poorly in low-texture regions with few events. Thus, synthesis-based and flow-based approaches are complementary. In this work, we introduce Time Lens, a novel method that leverages the advantages of both. We extensively evaluate our method on three synthetic and two real benchmarks where we show an up to 5.21 dB improvement in terms of PSNR over state-of-the-art frame-based and event-based methods. Finally, we release a new large-scale dataset in highly dynamic scenarios, aimed at pushing the limits of existing methods.

The timeline of computer vision research is marked with advances in learning and utilizing efficient contextual representations. Most of them, however, are targeted at improving model performance on a single downstream task. We consider a multi-task environment for dense prediction tasks, represented by a common backbone and independent task-specific heads. Our goal is to find the most efficient way to refine each task prediction by capturing cross-task contexts dependent on tasks’ relations. We explore various attention-based contexts, such as global and local, in the multi-task setting and analyze their behavior when applied to refine each task independently. Empirical findings confirm that different source-target task pairs benefit from different context types. To automate the selection process, we propose an Adaptive Task-Relational Context (ATRC) module, which samples the pool of all available contexts for each task pair using neural architecture search and outputs the optimal configuration for deployment. Our method achieves state-of-the-art performance on two important multi-task benchmarks, namely NYUD-v2 and PASCAL-Context. The proposed ATRC has a low computational toll and can be used as a drop-in refinement module for any supervised multi-task architecture.

Being able to learn dense semantic representations of images without supervision is an important problem in computer vision. However, despite its significance, this problem remains rather unexplored, with a few exceptions that considered unsupervised semantic segmentation on small-scale datasets with a narrow visual domain. In this paper, we make a first attempt to tackle the problem on datasets that have been traditionally utilized for the supervised case. To achieve this, we introduce a two-step framework that adopts a predetermined mid-level prior in a contrastive optimization objective to learn pixel embeddings. This marks a large deviation from existing works that relied on proxy tasks or end-to-end clustering. Additionally, we argue about the importance of having a prior that contains information about objects, or their parts, and discuss several possibilities to obtain such a prior in an unsupervised manner.Experimental evaluation shows that our method comes with key advantages over existing works. First, the learned pixel embeddings can be directly clustered in semantic groups using K-Means on PASCAL. Under the fully unsupervised setting, there is no precedent in solving the semantic segmentation task on such a challenging benchmark. Second, our representations can improve over strong baselines when transferred to new datasets, e.g. COCO and DAVIS. The code is available1.

Contrastive self-supervised learning has outperformed supervised pretraining on many downstream tasks like segmentation and object detection. However, current methods are still primarily applied to curated datasets like ImageNet. In this paper, we first study how biases in the dataset affect existing methods. Our results show that an approach like MoCo works surprisingly well across: (i) object- versus scene-centric, (ii) uniform versus long-tailed and (iii) general versus domain-specific datasets. Second, given the generality of the approach, we try to realize further gains with minor modifications. We show that learning additional invariances - through the use of multi-scale cropping, stronger augmentations and nearest neighbors - improves the representations. Finally, we observe that MoCo learns spatially structured representations when trained with a multi-crop strategy. The representations can be used for semantic segment retrieval and video instance segmentation without finetuning. Moreover, the results are on par with specialized models. We hope this work will serve as a useful study for other researchers.

We introduce T-Basis, a novel concept for a compact representation of a set of tensors, each of an arbitrary shape, which is often seen in Neural Networks. Each of the tensors in the set is modeled using Tensor Rings, though the concept applies to other Tensor Networks. Owing its name to the T-shape of nodes in diagram notation of Tensor Rings, T-Basis is simply a list of equally shaped three-dimensional tensors, used to represent Tensor Ring nodes. Such representation allows us to parameterize the tensor set with a small number of parameters (coefficients of the T-Basis tensors), scaling logarithmically with each tensor’s size in the set and linearly with the dimensionality of T-Basis. We evaluate the proposed approach on the task of neural network compression and demonstrate that it reaches high compression rates at acceptable performance drops. Finally, we analyze memory and operation requirements of the compressed networks and conclude that T-Basis networks are equally well suited for training and inference in resource-constrained environments and usage on the edge devices. Project website: obukhov.ai/tbasis.

Generating novel, yet realistic, images of persons is a challenging task due to the complex interplay between the different image factors, such as the foreground, background and pose information. In this work, we aim at generating such images based on a novel, two-stage reconstruction pipeline that learns a disentangled representation of the aforementioned image factors and generates novel person images at the same time. First, a multi-branched reconstruction network is proposed to disentangle and encode the three factors into embedding features, which are then combined to re-compose the input image itself. Second, three corresponding mapping functions are learned in an adversarial manner in order to map Gaussian noise to the learned embedding feature space, for each factor, respectively. Using the proposed framework, we can manipulate the foreground, background and pose of the input image, and also sample new embedding features to generate such targeted manipulations, that provide more control over the generation process. Experiments on the Market-1501 and Deepfashion datasets show that our model does not only generate realistic person images with new foregrounds, backgrounds and poses, but also manipulates the generated factors and interpolates the in-between states. Another set of experiments on Market-1501 shows that our model can also be beneficial for the person re-identification task1.

Modern cars are incorporating an increasing number of driver assist features, among which automatic lane keeping. The latter allows the car to properly position itself within the road lanes, which is also crucial for any subsequent lane departure or trajectory planning decision in fully autonomous cars. Traditional lane detection methods rely on a combination of highly-specialized, hand-crafted features and heuristics, usually followed by post-processing techniques, that are computationally expensive and prone to scalability due to road scene variations. More recent approaches leverage deep learning models, trained for pixel-wise lane segmentation, even when no markings are present in the image due to their big receptive field. Despite their advantages, these methods are limited to detecting a pre-defined, fixed number of lanes, e.g. ego-lanes, and can not cope with lane changes. In this paper, we go beyond the aforementioned limitations and propose to cast the lane detection problem as an instance segmentation problem - in which each lane forms its own instance - that can be trained end-to-end. To parametrize the segmented lane instances before fitting the lane, we further propose to apply a learned perspective transformation, conditioned on the image, in contrast to a fixed ”bird's-eye view” transformation. By doing so, we ensure a lane fitting which is robust against road plane changes, unlike existing approaches that rely on a fixed, predefined transformation. In summary, we propose a fast lane detection algorithm, running at 50 fps, which can handle a variable number of lanes and cope with lane changes. We verify our method on the tuSimple dataset and achieve competitive results.

In this paper, we present a method that estimates reflectance and illumination information from a single image depicting a single-material specular object from a given class under natural illumination. We follow a data-driven, learning-based approach trained on a very large dataset, but in contrast to earlier work we do not assume one or more components (shape, reflectance, or illumination) to be known. We propose a two-step approach, where we first estimate the object's reflectance map, and then further decompose it into reflectance and illumination. For the first step, we introduce a Convolutional Neural Network (CNN) that directly predicts a reflectance map from the input image itself, as well as an indirect scheme that uses additional supervision, first estimating surface orientation and afterwards inferring the reflectance map using a learning-based sparse data interpolation technique. For the second step, we suggest a CNN architecture to reconstruct both Phong reflectance parameters and high-resolution spherical illumination maps from the reflectance map. We also propose new datasets to train these CNNs. We demonstrate the effectiveness of our approach for both steps by extensive quantitative and qualitative evaluation in both synthetic and real data as well as through numerous applications, that show improvements over the state-of-the-art.

How much does a single image reveal about the environment it was taken in? In this paper, we investigate how much of that information can be retrieved from a foreground object, combined with the background (i.e. the visible part of the environment). Assuming it is not perfectly diffuse, the foreground object acts as a complexly shaped and far-from-perfect mirror An additional challenge is that its appearance confounds the light coming from the environment with the unknown materials it is made of. We propose a learning-based approach to predict the environment from multiple reflectance maps that are computed from approximate surface normals. The proposed method allows us to jointly model the statistics of environments and material properties. We train our system from synthesized training data, but demonstrate its applicability to real-world data. Interestingly, our analysis shows that the information obtained from objects made out of multiple materials often is complementary and leads to better performance.

Research on visual material recognition has traditionally been based on texture analysis. Whereas older work has focused on uncluttered scenes, more recent contributions allowed for material recognition 'in the wild'. Quite some objects have untextured surfaces, however. Especially man-made examples are legion. The most obvious cue to use in such cases would be reflection information. Yet, methods to that effect are lacking. It is clear that more than an estimate of a scalar albedo is needed, and a more complete reflectance model has to be derived. Rather than using an extensive lab setup, we propose a system that only requires the 3D shape of objects and a regular, commercial camera to capture their appearance in a single image, to perform material classification under unknown illumination. To this end, we rely on a Gaussian Process Latent Variable Model (GPLVM) with a discriminative prior to learn a low-dimensional manifold suitable for material classification of reflectance maps, i.e. from a 2D image of a singlematerial sphere under natural illumination. We evaluated our method based on experiments generated from synthetic and real-life data. Although recognizing materials without texture (or object recognition) is not a trivial problem, our method achieves about 75% recognition accuracy, about 27% higher than human performance.

The problem of estimating a full BRDF from partial observations has already been studied using either parametric or non-parametric approaches. The goal in each case is to best match this sparse set of input measurements. In this paper we address the problem of inferring higher order reflectance information starting from the minimal input of a single BRDF slice. We begin from the prototypical case of a homogeneous sphere, lit by a head-on light source, which only holds information about less than 0.001% of the whole BRDF domain. We propose a novel method to infer the higher dimensional properties of the material's BRDF, based on the statistical distribution of known material characteristics observed in real-life samples. We evaluated our method based on a large set of experiments generated from real-world BRDFs and newly measured materials. Although inferring higher dimensional BRDFs from such modest training is not a trivial problem, our method performs better than state-of-the-art parametric, semi-parametric and non-parametric approaches. Finally, we discuss interesting applications on material re-lighting, and flash-based photography.

In this work, we investigate the use of simple flash-based photography to capture an object's 3D shape and reflectance characteristics at the same time. The presented method is based on the principles of Structure from Motion (SfM) and Photometric Stereo (PS), yet, we make sure not to use more than readily-available consumer equipment, like a camera with flash. Starting from a SfM-generated mesh, we apply PS to refine both geometry and reflectance, where the latter is expressed in terms of data-driven Bidirectional Reflectance Distribution Function (BRDF) representations. We also introduce a novel approach to infer complete BRDFs starting from the sparsely sampled data-driven reflectance information captured with this setup. Our approach is experimentally validated by modeling several challenging objects, both synthetic and real.