ECCV 2020 Papers

We introduce a learning-based approach for room navigation using semantic maps. Our proposed architecture learns to predict top-down belief maps of regions that lie beyond the agent’s field of view while modeling architectural and stylistic regularities in houses. First, we train a model to generate amodal semantic top-down maps indicating beliefs of location, size, and shape of rooms by learning the underlying architectural patterns in houses. Next, we use these maps to predict a point that lies in the target room and train a policy to navigate to the point. We empirically demonstrate that by predicting semantic maps, the model learns common correlations found in houses and generalizes to novel environments. We also demonstrate that reducing the task of room navigation to point navigation improves the performance further. We will make our code publicly available and hope our work paves the way for further research in this space. "

Self-driving cars need to understand 3D scenes efficiently and accurately in order to drive safely. Given the limited hardware resources, existing 3D perception models are not able to recognize small instances (e.g., pedestrians, cyclists) very well due to the low-resolution voxelization and aggressive downsampling. To this end, we propose Sparse Point-Voxel Convolution (SPVConv), a lightweight 3D module that equips the vanilla Sparse Convolution with the high-resolution point-based branch. With negligible overhead, this point-based branch is able to preserve the fine details even from large outdoor scenes. To explore the spectrum of efficient 3D models, we first define a flexible architecture design space based on SPVConv, and we then present 3D Neural Architecture Search (3D-NAS) to search the optimal network architecture over this diverse design space efficiently and effectively. Experimental results validate that the resulting SPVNAS model is fast and accurate: it outperforms the state-of-the-art MinkowskiNet by 3.3%, ranking 1st on the competitive SemanticKITTI leaderboard. It also achieves 8-24× computation reduction and 3-35× measured speedup over MinkowskiNet and KPConv with higher accuracy. Finally, we transfer our method to 3D object detection, and it achieves consistent improvements over the one-stage detection baseline on KITTI."

Emergent hardwares can support mixed precision CNN models inference that assign different bitwidths for different layers. Learning to find an optimal mixed precision model that can preserve accuracy and satisfy the specific constraints on model size and computation is extremely challenge due to the difficult in training a mixed precision model and the huge space of all possible bit quantizations. In this paper, we propose a novel soft Barrier Penalty based NAS (BP-NAS) for mixed precision quantization, which ensures all the searched models are inside the valid domain defined by the complexity constraint, thus could return an optimal model under the given constraint by conducting search only one time. The proposed soft Barrier Penalty is differentiable and can impose very large losses to those models outside the valid domain while almost no punishment for models inside the valid domain, thus constraining the search only in the feasible domain. In addition, a differentiable Prob-1 regularizer is proposed to ensure learning with NAS is reasonable. A distribution reshaping training strategy is also used to make training more stable. BP-NAS sets new state of the arts on both classification (Cifar-10, ImageNet) and detection (COCO), surpassing all the efficient mixed precision methods designed manually and automatically. Particularly, BP-NAS achieves higher mAP (up to 2.7% mAP improvement) together with lower bit computation cost compared with the existing best mixed precision model on COCO detection."

Manually labeling video datasets for segmentation tasks is extremely time consuming. We introduce ScribbleBox, an interactive framework for annotating object instances with masks in videos with a significant boost in efficiency. In particular, we split annotation into two steps: annotating objects with tracked boxes, and labeling masks inside these tracks. We introduce automation and interaction in both steps. Box tracks are annotated efficiently by approximating the trajectory using a parametric curve with a small number of control points which the annotator can interactively correct. Our approach tolerates a modest amount of noise in box placements, thus typically requiring only a few clicks to annotate a track to a sufficient accuracy. Segmentation masks are corrected via scribbles which are propagated through time. We show significant performance gains in annotation efficiency over past work. We show that our ScribbleBox approach reaches 88.92% J&F on DAVIS2017 with an average of 9.14 clicks per box track, and only 4 frames requiring scribble annotation in a video of 65.3 frames on average."

Sketch-based image retrieval (SBIR) has been a popular research topic in recent years. Existing works concentrate on mapping the visual information of sketches and images to a semantic space at the object level. In this paper, for the first time, we study the fine-grained scene-level SBIR problem which aims at retrieving scene images satisfying the user's specific requirements via a freehand scene sketch. We propose a graph embedding based method to learn the similarity measurement between images and scene sketches, which models the multi-modal information, including the size and appearance of objects as well as their layout information, in an effective manner. To evaluate our approach, we collect a dataset based on SketchyCOCO and extend the dataset using Coco-stuff. Comprehensive experiments demonstrate the significant potential of the proposed approach on the application of fine-grained scene-level image retrieval."

We present a novel approach to reconstructing lightweight, CAD-based representations of scanned 3D environments from commodity RGB-D sensors. Our key idea is to jointly optimize for both CAD model alignments as well as layout estimations of the scanned scene, explicitly modeling inter-relationships between objects-to-objects and objects-to-layout. Since object arrangement and scene layout are intrinsically coupled, we show that treating the problem jointly significantly helps to produce globally-consistent representations of a scene. Object CAD models are aligned to the scene by establishing dense correspondences between geometry, and we introduce a hierarchical layout prediction approach to estimate layout planes from corners and edges of the scene. To this end, we propose a message-passing graph neural network to model the inter-relationships between objects and layout, guiding generation of a globally object alignment in a scene. By considering the global scene layout, we achieve significantly improved CAD alignments compared to state-of-the-art methods, improving from 41.83% to 58.41% alignment accuracy on SUNCG and from 50.05% to 61.24% on ScanNet, respectively. The resulting CAD-based representations makes our method well-suited for applications in content creation such as augmented- or virtual reality."

Low-resolution text images are often seen in natural scenes such as documents captured by mobile phones. Recognizing low-resolution text images is challenging because they lose detailed content information, leading to poor recognition accuracy. An intuitive solution is to introduce super-resolution (SR) techniques as pre-processing. However, previous single image super-resolution (SISR) methods are trained on synthetic low resolution images (e.g.Bicubic down-sampling), which is simple and not suitable for real low-resolution text recognition. To this end, we propose a real scene text SR dataset, termed TextZoom. It contains paired real low-resolution and high-resolution images which are captured by cameras with different focal length in the wild. It is more authentic and challenging than synthetic data, as shown in Fig. 1. We argue improving the recognition accuracy is the ultimate goal for SceneText SR. In this purpose, a new Text Super-Resolution Network, termed TSRN, with three novel modules is developed. (1) A sequential residual block is proposed to extract the sequential information of the text images. (2) A boundary-aware loss is designed to sharpen the character boundaries.(3) A central alignment module is proposed to relieve the misalignment problem in TextZoom. Extensive experiments on TextZoom demonstrate that our TSRN largely improves the recognition accuracy by over 13%of CRNN, and by nearly 9.0% of ASTER and MORAN compared to synthetic SR data. Furthermore, our TSRN clearly outperforms 7 state-of-the-art SR methods in boosting the recognition accuracy of LR images in TextZoom. For example, it outperforms LapSRN by over 5% and 8%on the recognition accuracy of ASTER and CRNN. Our results suggest that low-resolution text recognition in the wild is far from being solved, thus more research effort is needed"

We introduce the new task of 3D object localization in RGB-D scans using natural language descriptions. As input, we assume a point cloud of a scanned 3D scene along with a free-form description of a specified target object. To address this task, we propose ScanRefer, where the core idea is to learn a fused descriptor from 3D object proposals and encoded sentence embeddings. This learned descriptor then correlates the language expressions with the underlying geometric features of the 3D scan and facilitates the regression of the 3D bounding box of the target object. In order to train and benchmark our method, we introduce a new ScanRefer dataset, containing 46,173 descriptions of 9,943 objects from 703 ScanNet scenes. ScanRefer is the first large-scale effort to perform object localization via natural language expression directly in 3D."

Inferring road graphs from satellite imagery is a challenging computer vision task. Prior solutions fall into two categories: (1) pixel-wise segmentation-based approaches, which predict whether each pixel is on a road, and (2) graph-based approaches, which predict the road graph iteratively. We find that these two approaches have complementary strengths while suffering from their own inherent limitations.

Existing methods for instance segmentation in videos typically involve multi-stage pipelines that follow the tracking-by detection paradigm and model a video clip as a sequence of images. Multiple networks are used to detect objects in individual frames, and then associate these detections over time. Hence, these methods are often non-end-to-end trainable and highly tailored to specific tasks. In this paper, we propose a different approach that is well-suited to a variety of tasks involving instance segmentation in videos. In particular, we model a video clip as a single 3D spatio-temporal volume, and propose a novel approach that segments and tracks instances across space and time in a single stage. Our problem formulation is centered around the idea of spatio-temporal embeddings which are trained to cluster pixels belonging to a specific object instance over an entire video clip. To this end, we introduce (i) novel mixing functions that enhance the feature representation of spatio-temporal embeddings, and (ii) a single-stage, proposal-free network that can reason about temporal context. Our network is trained end-to-end to learn spatio-temporal embeddings as well as parameters required to cluster these embeddings, thus simplifying inference. Our method achieves state-of-the-art results across multiple datasets and tasks."

The SMPL body model is widely used for the estimation, synthesis, and analysis of 3D human pose and shape. While popular, we show that SMPL has several limitations and introduce STAR, which is quantitatively and qualitatively superior to SMPL. First, SMPL has a huge number of parameters resulting from its use of global blend shapes. These dense pose-corrective offsets relate every vertex on the mesh to all the joints in the kinematic tree, capturing spurious long-range cor- relations. To address this, we define per-joint pose correctives and learn the subset of mesh vertices that are influenced by each joint movement. This sparse formulation results in more realistic deformations and significantly reduces the number of model parameters to 20% of SMPL. When trained on the same data as SMPL, STAR generalizes better despite having many fewer parameters. Second, SMPL factors pose-dependent deformations from body shape while, in reality, people with different shapes deform differently. Consequently, we learn shape-dependent pose- corrective blendshapes that depend on both body pose and BMI. Third, we show that the shape space of SMPL is not rich enough to capture the variation in the human population. We address this by training STAR with an additional 10,000 scans of male and female subjects, and show that this results in better model generalization. STAR is compact, generalizes better to new bodies and is a drop-in replacement for SMPL. STAR is publicly available for research purposes at http://star.is.tue.mpg.de."

Multi-class 3D object detection aims to classify and localize objects of multiple categories from point clouds. Due to the nature of point clouds, i.e. unstructured, sparse and noisy, some features benefit-ting multi-class discrimination are underexploited, such as shape information. In this paper, we propose a novel 3D shape signature to explore the shape information from point clouds. By incorporating operations of symmetry, convex hull and chebyshev fitting, the proposed shape sig-nature is not only compact and effective but also robust to the noise, which serves as a soft constraint to improve the feature capability of multi-class discrimination. Based on the proposed shape signature, we develop the shape signature networks (SSN) for 3D object detection, which consist of pyramid feature encoding part, shape-aware grouping heads and explicit shape encoding objective. Experiments show that the proposed method performs remarkably better than existing methods on two large-scale datasets. Furthermore, our shape signature could act as a plug-and-play component and ablation study shows its effectiveness and good scalability"

Existing facial attribute editing methods typically employ an encoder-decoder architecture where the attribute information is expressed as a conditional one-hot vector spatially concatenated with the image or intermediate feature maps. However, such operations only learn the local semantic mapping but ignore global facial statistics. In this work, we focus on solving this issue by editing the channel-wise global information denoted as the style feature. We develop a style skip connection based generative adversarial network, referred to as SSCGAN which enables accurate facial attribute manipulation. Specifically, we inject the target attribute information into multiple style skip connection paths between the encoder and decoder. Each connection extracts the style feature of the latent feature maps in the encoder and then performs a residual learning based mapping function in the global information space guided by the target attributes. In the following, the adjusted style feature will be utilized as the conditional information for instance normalization to transform the corresponding latent feature maps in the decoder. In addition, to avoid the vanishing of spatial details ( extit{e.g.} hairstyle or pupil locations), we further introduce the skip connection based spatial information transfer module. Through the global-wise style and local-wise spatial information manipulation, the proposed method can produce better results in terms of attribute generation accuracy and image quality. Experimental results demonstrate the proposed algorithm performs favorably against the state-of-the-art methods."

Human poses that are rare or unseen in a training set are challenging for a network to predict. Similar to the long-tailed distribution problem in visual recognition, the small number of examples for such poses limits the ability of networks to model them. Interestingly, local pose distributions suffer less from the long-tail problem, i.e., local joint configurations within a rare pose may appear within other poses in the training set, making them less rare. We propose to take advantage of this fact for better generalization to rare and unseen poses. To be specific, our method splits the body into local regions and processes them in separate network branches, utilizing the property that a joint’s position depends mainly on the joints within its local body region. Global coherence is maintained by recombining the global context from the rest of the body into each branch as a low-dimensional vector. With the reduced dimensionality of less relevant body areas, the training set distribution within network branches more closely reflects the statistics of local poses instead of global body poses, without sacrificing information important for joint inference. The proposed split-and-recombine approach, called SRNet, can be easily adapted to both single-image and temporal models, and it leads to appreciable improvements in the prediction of rare and unseen poses."

Super-resolution is an ill-posed problem, since it allows for multiple predictions for a given low-resolution image. This fundamental fact is largely ignored by state-of-the-art deep learning based approaches. These methods instead train a deterministic mapping using combinations of reconstruction and adversarial losses. In this work, we therefore propose SRFlow: a normalizing flow based super-resolution method capable of learning the conditional distribution of the output given the low-resolution input. Our model is trained in a principled manner using a single loss, namely the negative log-likelihood. SRFlow therefore directly accounts for the ill-posed nature of the problem, and learns to predict diverse photo-realistic high-resolution images. Moreover, we utilize the strong image posterior learned by SRFlow to design flexible image manipulation techniques, capable of enhancing super-resolved images by, e.g., transferring content from other images. We perform extensive experiments on faces, as well as on super-resolution in general. SRFlow outperforms state-of-the-art GAN-based approaches in terms of both PSNR and perceptual quality metrics, while allowing for diversity through the exploration of the space of super-resolved solutions."

The sparsity of point clouds limits deep learning models on capturing long-range dependencies, which makes features extracted by the models ambiguous. In point cloud object detection, ambiguous features make it hard for detectors to locate object centers and finally lead to bad detection results. In this work, we propose Selective Point clOud voTing (SPOT) module, a simple effective component that can be easily trained end-to-end in point cloud object detectors to solve this problem. Inspired by probabilistic Hough voting, SPOT incorporates an attention mechanism that helps detectors focus on less ambiguous features and preserves their diversity of mapping to multiple object centers. For evaluating our module, we implement SPOT on advanced baseline detectors and test on two benchmark datasets of clutter indoor scenes, ScanNet and SUN RGB-D. Baselines enhanced by our module can stably improve results in agreement by a large margin and achieve new state-or-the-art detection, especially under more strict evaluation metric that adopts larger IoU threshold, implying our module is the key leading to high-quality object detection in point clouds."

Although significant progress achieved, multi-label classification is still challenging due to the complexity of correlations among different labels. Furthermore, modeling the relationships between input and some (dull) classes further increases the difficulty of accurately predicting all possible labels. In this work, we propose to select a small subset of labels as landmarks which are easy to predict according to input (predictable) and can well recover the other possible labels (representative). Different from existing methods which separate the landmark selection and landmark prediction in the 2-step manner, the proposed algorithm, termed Selecting Predictable Landmarks for MultiLabel Learning (SPL-MLL), jointly conducts landmark selection, landmark prediction, and label recovery in a unified framework, to ensure both the representativeness and predictableness for selected landmarks. We employ the Alternating Direction Method (ADM) to solve our problem. Empirical studies on real-world datasets show that our method achieves superior classification performance over other state-of-the-art methods."

Adversarial attacks of deep neural networks have been intensively studied on image, audio, natural language, patch, and pixel classification tasks. Nevertheless, as a typical, while important real-world application, the adversarial attacks of online video object tracking that traces an object's moving trajectory instead of its category are rarely explored. In this paper, we identify a new task for the adversarial attack to visual tracking: online generating imperceptible perturbations that mislead trackers along with an incorrect (Untargeted Attack, UA) or specified trajectory (Targeted Attack, TA). To this end, we first propose a spatial-aware basic attack by adapting existing attack methods, \ie, FGSM, BIM, and C\&W, and comprehensively analyze the attacking performance. We identify that online object tracking poses two new challenges: 1) it is difficult to generate imperceptible perturbations that can transfer across frames, and 2) real-time trackers require the attack to satisfy a certain level of efficiency. To address these challenges, we further propose the spatial-aware online incremental attack (a.k.a. SPARK) that performs spatial-temporal sparse incremental perturbations online and makes the adversarial attack less perceptible. In addition, as an optimization-based method, SPARK quickly converges to very small losses within several iterations by considering historical incremental perturbations, making it much more efficient than basic attacks. The in-depth evaluation of state-of-the-art trackers (i.e., SiamRPN++ with AlexNet, MobileNetv2, and ResNet-50, and SiamDW) on OTB100, VOT2018, UAV123, and LaSOT demonstrates the effectiveness and transferability of SPARK in misleading the trackers under both UA and TA with minor perturbations."

Tehchniques for manipulating images are advancing rapidly; while these are helpful for many useful tasks, they also pose a threat to society with their ability to create believable misinformation. We present a novel, Spatial Pyramid Attention Network (SPAN) for detection and localization of multiple types of image manipulations. The proposed architecture efficiently and effecively models the relationship between image patches at multiple scales by constructing a pyramid of local self-attention blocks. The design includes a novel position projection to encode the spatial positions of the patches. SPAN is trained on a synthetic dataset but can also be fine tuned for specific datasets; The proposed method shows significant gains in performance on standard datasets over previous state-of-art methods. "