To get clear street-view and photo-realistic simulation in autonomous driving, we present an automatic video inpainting algorithm that can remove traffic agents from videos and synthesize missing regions with the guidance of depth/point cloud. By building a dense 3D map from stitched point clouds, frames within a video are geometrically correlated via this common 3D map. In order to fill a target inpainting area in a frame, it is straightforward to transform pixels from other frames into the current one with correct occlusion. Furthermore, we are able to fuse multiple videos through 3D point cloud registration, making it possible to inpaint a target video with multiple source videos. The motivation is to solve the long-time occlusion problem where an occluded area has never been visible in the entire video. This happens quite often for street view videos when there is a vehicle in the front. In this case, we can capture the scene a second time when the desired area becomes visible and do video fusion inpainting. To our knowledge, we are the first to fuse multiple videos for video inpainting. In order to evaluate our method, we collected 5 hour synchronized lidar and camera data from autonomous driving cars in the urban roads. To address long-time occlusion problems, we collect data for the same road in different days and times. The experimental results show that the proposed approach outperforms the state-of-the-art approaches for all the criteria, especially the RMSE (Root Mean Squared Error) has been reduced by about %13.
论文检索
输入标题、作者或关键词，从 1,358 篇学术成果中精准定位
The problem of catastrophic forgetting occurs in deep learning models trained on multiple databases in a sequential manner. Recently, generative replay mechanisms (GRM), have been proposed to reproduce previously learned knowledge aiming to reduce the forgetting. However, such approaches lack an appropriate inference model and therefore can not provide latent representations of data. In this paper, we propose a novel lifelong learning approach, namely the Lifelong VAEGAN (L-VAEGAN), which not only induces a powerful generative replay network but also learns meaningful latent representations, benefiting representation learning. L-VAEGAN can allow to automatically embed the information associated with different domains into several clusters in the latent space, while also capturing semantically meaningful shared latent variables, across different data domains. The proposed model supports many downstream tasks that traditional generative replay methods can not, including interpolation and inference across different data domains.
We propose a Dynamic Directed Graph Convolutional Network (DDGCN) to model spatial and temporal features of human actions from their skeletal representations. The DDGCN consists of three new feature modeling modules: (1) Dynamic Convolutional Sampling (DCS), (2) Dynamic Convolutional Weight (DCW) assignment, and (3) Directed Graph Spatial-Temporal (DGST) feature extraction. Comprehensive experiments show that the DDGCN outperforms existing state-of-the-art action recognition approaches in various testing datasets. Our source code and model will be released at http://www.ece.lsu.edu/xinli/ActionModeling/index.html .
We propose a novel algorithm for the fitting of 3D human shape to images. Combining the accuracy and refinement capabilities of iterative gradient-based optimization techniques with the robustness of deep neural networks, we propose a gradient descent algorithm that leverages a neural network to predict the parameter update rule for each iteration. This per-parameter and state-aware update guides the optimizer towards a good solution in very few steps, converging in typically few steps. During training our approach only requires MoCap data of human poses, parametrized via SMPL. From this data the network learns a subspace of valid poses and shapes in which optimization is performed much more efficiently. The approach does not require any hard to acquire image-to-3D correspondences. At test time we only optimize the 2D joint re-projection error without the need for any further priors or regularization terms. We show empirically that this algorithm is fast (avg. 120ms convergence), robust to initialization and dataset, and achieves state-of-the-art results on public evaluation datasets including the challenging 3DPW in-the-wild benchmark (improvement over SMPLify ($45\%$) and also approaches using image-to-3D correspondences).
Image retrieval is the problem of searching an image database for items that are similar to a query image. To address this task, two main types of image representations have been studied: global and local image features. In this work, our key contribution is to unify global and local features into a single deep model, enabling accurate retrieval with efficient feature extraction. We refer to the new model as DELG, standing for DEep Local and Global features. We leverage lessons from recent feature learning work and propose a model that combines generalized mean pooling for global features and attentive selection for local features. The entire network can be learned end-to-end by carefully balancing the gradient flow between two heads -- requiring only image-level labels. We also introduce an autoencoder-based dimensionality reduction technique for local features, which is integrated into the model, improving training efficiency and matching performance. Comprehensive experiments show that our model achieves state-of-the-art image retrieval on the Revisited Oxford and Paris datasets, and state-of-the-art single-model instance-level recognition on the Google Landmarks dataset v2. Code and models are available at https://github.com/tensorflow/models/tree/master/research/delf .
Learning predictive models from interaction with the world allows an agent, such as a robot, to learn about how the world works, and then use this learned model to plan coordinated sequences of actions to bring about desired outcomes. However, learning a model that captures the dynamics of complex skills represents a major challenge: if the agent needs a good model to perform these skills, it might never be able to collect the experience on its own that is required to learn these delicate and complex behaviors. Instead, we can imagine augmenting the training set with observational data of other agents, such as humans. Such data is likely more plentiful, but cannot always be combined with data from the original agent. For example, videos of humans might show a robot how to use a tool, but (i) are not annotated with suitable robot actions, and (ii) contain a systematic distributional shift due to the embodiment differences between humans and robots. We address the first challenge by formulating the corresponding graphical model and treating the action as an observed variable for the interaction data and an unobserved variable for the observation data, and the second challenge by using a domain-dependent prior. In addition to interaction data, our method is able to leverage videos of passive observations in a driving dataset and a dataset of robotic manipulation videos to improve video prediction performance. We also demonstrate that, in a real-world tabletop robotic manipulation setting, we are able to significantly improve control performance by learning a model from both robot data and observations of humans.
Multi-task networks are commonly utilized to alleviate the need for a large number of highly specialized single-task networks. However, two common challenges in developing multi-task models are often overlooked in literature. First, enabling the model to be inherently incremental, continuously incorporating information from new tasks without forgetting the previously learned ones (incremental learning). Second, eliminating adverse interactions amongst tasks, which has been shown to significantly degrade the single-task performance in a multi-task setup (task interference). In this paper, we show that both can be achieved simply by reparameterizing the convolutions of standard neural network architectures into a non-trainable shared part (filter bank) and task-specific parts (modulators), where each modulator has a fraction of the filter bank parameters. Thus, our reparameterization enables the model to learn new tasks without adversely affecting the performance of existing ones. The results of our ablation study attest the efficacy of the proposed reparameterization. Moreover, our method achieves state-of-the-art on two challenging multi-task learning benchmarks, PASCAL-Context and NYUD, and also demonstrates superior incremental learning capability as compared to its close competitors. The code and models are made publicly available.
Existing gradient-based attribution-map methods rely on hand-crafted propagation rules for the non-linear/activation layers during the backward pass, so as to produce gradients of the input and then the attribution map. Despite the promising results achieved, such methods are sensitive to the non-informative high-frequency components and lack adaptability for various models and samples. In this paper, we propose a dedicated method to generate attribution maps that allow us to learn the propagation rules automatically, overcoming the flaws of the hand-crafted ones. Specifically, we introduce a learnable plugin module, which enables adaptive propagation rules for each pixel, to the non-linear layers during the backward pass for mask generating. The masked input image is then fed into the model again to obtain new output that can be used as a guidance when combined with the original one. The introduced learnable module can be trained under any auto-grad framework with higher-order differential support. As demonstrated on several datasets and network architectures, the proposed method yields state-of-the-art results and gives cleaner and more visually plausible attribution maps.
AssembleNet++: Assembling Modality Representations via Attention Connections - Supplementary Material -
PDF ↗We create a family of powerful video models which are able to: (i) learn interactions between semantic object information and raw appearance and motion features, and (ii) deploy attention in order to better learn the importance of features at each convolutional block of the network. A new network component named peer-attention is introduced, which dynamically learns the attention weights using another block or input modality. Even without pre-training, our models outperform the previous work on standard public activity recognition datasets with continuous videos, establishing new state-of-the-art. We also confirm that our findings of having neural connections from the object modality and the use of peer-attention is generally applicable for different existing architectures, improving their performances. We name our model explicitly as AssembleNet++. The code will be available at: https://sites.google.com/corp/view/assemblenet/
Semantic image understanding is a challenging topic in computer vision. It requires to detect all objects in an image, but also to identify all the relations between them. Detected objects, their labels and the discovered relations can be used to construct a scene graph which provides an abstract semantic interpretation of an image. In previous works, relations were identified by solving an assignment problem formulated as Mixed-Integer Linear Programs. In this work, we interpret that formulation as Ordinary Differential Equation (ODE). The proposed architecture performs scene graph inference by solving a neural variant of an ODE by end-to-end learning. It achieves state-of-the-art results on all three benchmark tasks: scene graph generation (SGGen), classification (SGCls) and visual relationship detection (PredCls) on Visual Genome benchmark.
The 2D virtual try-on task has recently attracted a great interest from the research community, for its direct potential applications in online shopping as well as for its inherent and non-addressed scientific challenges. This task requires fitting an in-shop cloth image on the image of a person, which is highly challenging because it involves cloth warping, image compositing, and synthesizing. Casting virtual try-on into a supervised task faces a difficulty: available datasets are composed of pairs of pictures (cloth, person wearing the cloth). Thus, we have no access to ground-truth when the cloth on the person changes. State-of-the-art models solve this by masking the cloth information on the person with both a human parser and a pose estimator. Then, image synthesis modules are trained to reconstruct the person image from a masked person image and a cloth image. This procedure has several caveats: firstly, human parsers are prone to errors; secondly, it is a costly pre-processing step, which also has to be applied at inference time; finally, it makes the task harder than it is since the mask covers information that should be kept such as hands or accessories. In this paper, we propose a novel student-teacher paradigm where the teacher is trained in the standard way (reconstruction) before guiding the student to focus on the initial task (changing the cloth). The student additionally learns from an adversarial loss, which pushes it to follow the distribution of the real images. Consequently, the student exploits information that is masked to the teacher. A student trained without the adversarial loss would not use this information. Also, getting rid of both human parser and pose estimator at inference time allows obtaining a real-time virtual try-on.
Temporal Activity Localization via Language (TALL) in video is a recently proposed challenging vision task, and tackling it requires fine-grained understanding of the video content, however, this is overlooked by most of the existing works. In this paper, we propose a novel TALL method which builds a Hierarchical Visual-Textual Graph to model interactions between the objects and words as well as among the objects to jointly understand the video contents and the language. We also design a convolutional network with cross-channel communication mechanism to further encourage the information passing between the visual and textual modalities. Finally, we propose a loss function that enforces alignment of the visual representation of the localized activity and the sentence representation, so that the model can predict more accurate temporal boundaries. We evaluated our proposed method on two popular benchmark datasets: Charades-STA and ActivityNet Captions, and achieved state-of-the-art performances on both datasets. Code is available at \url{https://github.com/forwchen/HVTG}.
Self-Supervised Monocular Depth Estimation: Solving the Dynamic Object Problem by Semantic Guidance
PDF ↗Self-supervised monocular depth estimation presents a powerful method to obtain 3D scene information from single camera images, which is trainable on arbitrary image sequences without requiring depth labels, e.g., from a LiDAR sensor. In this work we present a new self-supervised semantically-guided depth estimation (SGDepth) method to deal with moving dynamic-class (DC) objects, such as moving cars and pedestrians, which violate the static-world assumptions typically made during training of such models. Specifically, we propose (i) mutually beneficial cross-domain training of (supervised) semantic segmentation and self-supervised depth estimation with task-specific network heads, (ii) a semantic masking scheme providing guidance to prevent moving DC objects from contaminating the photometric loss, and (iii) a detection method for frames with non-moving DC objects, from which the depth of DC objects can be learned. We demonstrate the performance of our method on several benchmarks, in particular on the Eigen split, where we exceed all baselines without test-time refinement.
To learn a reliable people counter from crowd images, head center annotations are normally required. Annotating head centers is however a laborious and tedious process in dense crowds. In this paper, we present an active learning framework which enables accurate crowd counting with limited supervision: given a small labeling budget, instead of randomly selecting images to annotate, we first introduce an active labeling strategy to annotate the most informative images in the dataset and learn the counting model upon them. The process is repeated such that in every cycle we select the samples that are diverse in crowd density and dissimilar to previous selections. In the last cycle when the labeling budget is met, the large amount of unlabeled data are also utilized: a distribution classifier is introduced to align the labeled data with unlabeled data; furthermore, we propose to mix up the distribution labels and latent representations of data in the network to particularly improve the distribution alignment in-between training samples. We follow the popular density estimation pipeline for crowd counting. Extensive experiments are conducted on standard benchmarks i.e. ShanghaiTech, UCF CC 50, MAll, TRANCOS, and DCC. By annotating limited number of images (e.g. 10% of the dataset), our method reaches levels of performance not far from the state of the art which utilize full annotations of the dataset.
InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image
PDF ↗Analysis of hand-hand interactions is a crucial step towards better understanding human behavior. However, most researches in 3D hand pose estimation have focused on the isolated single hand case. Therefore, we firstly propose (1) a large-scale dataset, InterHand2.6M, and (2) a baseline network, InterNet, for 3D interacting hand pose estimation from a single RGB image. The proposed InterHand2.6M consists of extbf{2.6M labeled single and interacting hand frames} under various poses from multiple subjects. Our InterNet simultaneously performs 3D single and interacting hand pose estimation. In our experiments, we demonstrate big gains in 3D interacting hand pose estimation accuracy when leveraging the interacting hand data in InterHand2.6M. We also report the accuracy of InterNet on InterHand2.6M, which serves as a strong baseline for this new dataset. Finally, we show 3D interacting hand pose estimation results from general images. Our code and dataset are available ootnote{\url{https://mks0601.github.io/InterHand2.6M/}}.
This paper proposes a set of rules to revise various neural networks for 3D point cloud processing to rotation-equivariant quaternion neural networks (REQNNs). We find that when a neural network uses quaternion features, the network feature naturally has the rotation-equivariance property. Rotation equivariance means that applying a specific rotation transformation to the input point cloud is equivalent to applying the same rotation transformation to all intermediate-layer quaternion features. Besides, the REQNN also ensures that the intermediate-layer features are invariant to the permutation of input points. Compared with the original neural network, the REQNN exhibits higher rotation robustness.
Motion segmentation, i.e., the problem of clustering data in multiple images based on different 3D motions, is an important task for reconstructing and understanding dynamic scenes. In this paper we address motion segmentation in multiple images by combining partial results coming from triplets of images, which are obtained by fitting a number of trifocal tensors to correspondences. We exploit the fact that the trifocal tensor is a stronger model than the fundamental matrix, as it provides fewer but more reliable matches over three images than fundamental matrices provide over the two. We also consider an alternative solution which merges partial results coming from both triplets and pairs of images, showing the strength of three-frame segmentation in a combination with two-frame segmentation. Our real experiments on standard as well as new datasets demonstrate the superior accuracy of the proposed approaches when compared to previous techniques.
Recent generative adversarial network (GAN) based methods (e.g., CycleGAN) are prone to fail at preserving image-objects in image-to-image translation, which reduces their practicality on tasks such as domain adaptation. Some frameworks have been proposed to adopt a segmentation network as the auxiliary regularization to prevent the content distortion. However, all of them require extra pixel-wise annotations, which is difficult to fulfill in practical applications. In this paper, we propose a novel GAN (namely OP-GAN) to address the problem, which involves a self-supervised module to enforce the image content consistency during image-to-image translations without any extra annotations. We evaluate the proposed OP-GAN on three publicly available datasets. The experimental results demonstrate that our OP-GAN can yield visually plausible translated images and significantly improve the semantic segmentation accuracy in different domain adaptation scenarios with off-the-shelf deep learning networks such as PSPNet and U-Net.
The introduction of non-local blocks to the traditional CNN architecture enhances its performance for various computer vision tasks by improving its capabilities of capturing long range dependencies. However, the usage of non-local blocks may also introduce new threats to computer vision systems. Therefore, it is important to study the threats caused by non-local blocks before directly applying them to commercial systems. In this paper, two new threats named disappearing attack and appearing attack against object detectors with a non-local block are investigated. The former aims at misleading an object detector with a non-local block such that it is unable to detect a target object category while the latter aims at misleading the object detector such that it detects a predefined object category, which is not present in images. Different from the existing attacks against object detectors, these threats are able to be performed in long range cases. This means that the target object and the universal adversarial patches learned from the proposed algorithms can have long distance between them. To examine the threats, digital and physical experiments are conducted on Faster R-CNN with a non-local block and 6331 images from 56 videos. The experiments show that the universal patches are able to mislead the detector with greater probabilities. To explain the threats from non-local blocks, the reception fields of CNN models with and without non-local blocks are studied empirically and theoretically.
In this paper, we propose a Rotation-robust Intersection over Union ($ extit{RIoU}$) for 3D object detection, which aims to jointly learn the overlap of rotated bounding boxes. In most existing 3D object detection methods, the norm-based loss is adopted to individually regress the parameters of bounding boxes, which may suffer from the loss-metric mismatch due to the scaling problem. Motivated by the IoU loss in the axis-aligned 2D object detection which is invariant to the scale, our method jointly optimizes the parameters via the $ extit{RIoU}$ loss. To tackle the uncertainty of convex caused by rotation, a projection operation is defined to estimate the intersection area. The calculation process of $ extit{RIoU}$ and its loss function is robust to the rotation condition and feasible for back-propagation, which only comprises basic numerical operations. By incorporating the $ extit{RIoU}$ loss with the conventional norm-based loss function, we enforce the network to directly optimize the $ extit{RIoU}$ . Experimental results on the KITTI and nuScenes datasets validate the effectiveness of our proposed method. Moreover, we show that our method is suitable for the detection task of 2D rotated objects, such as text boxes and cluttered targets in the aerial images.