Search

Using generative models to synthesize new data has become a de-facto standard in autonomous driving to address the data scarcity issue. Though existing approaches are able to boost perception models, we discover that these approaches fail to improve the performance of planning of end-to-end autonomous driving models as the generated videos are usually less than 8 frames and the spatial and temporal inconsistencies are not negligible. To this end, we propose Delphi, a novel diffusion-based long video generation method with a shared noise modeling mechanism across the multi-views to increase spatial consistency, and a feature-aligned module to achieves both precise controllability and temporal consistency. Our method can generate up to 40 frames of video without loss of consistency which is about 5 times longer compared with state-of-the-art methods. Instead of randomly generating new data, we further design a sampling policy to let Delphi generate new data that are similar to those failure cases to improve the sample efficiency. This is achieved by building a failure-case driven framework with the help of pre-trained visual language models. Our extensive experiment demonstrates that our Delphi generates a higher quality of long videos surpassing previous state-of-the-art methods. Consequentially, with only generating 4% of the training dataset size, our framework is able to go beyond perception and prediction tasks, for the first time to the best of our knowledge, boost the planning performance of the end-to-end autonomous driving model by a margin of 25%., Comment: Project Page: https://westlake-autolab.github.io/delphi.github.io/, 8 figures

Visual object tracking plays a critical role in visual-based autonomous systems, as it aims to estimate the position and size of the object of interest within a live video. Despite significant progress made in this field, state-of-the-art (SOTA) trackers often fail when faced with adversarial perturbations in the incoming frames. This can lead to significant robustness and security issues when these trackers are deployed in the real world. To achieve high accuracy on both clean and adversarial data, we propose building a spatial-temporal continuous representation using the semantic text guidance of the object of interest. This novel continuous representation enables us to reconstruct incoming frames to maintain semantic and appearance consistency with the object of interest and its clean counterparts. As a result, our proposed method successfully defends against different SOTA adversarial tracking attacks while maintaining high accuracy on clean data. In particular, our method significantly increases tracking accuracy under adversarial attacks with around 90% relative improvement on UAV123, which is even higher than the accuracy on clean data.

We advocate the idea of the natural-language-driven(NLD) simulation to efficiently produce the object interactions between multiple objects in the virtual road scenes, for teaching and testing the autonomous driving systems that should take quick action to avoid collision with obstacles with unpredictable motions. The NLD simulation allows the brief natural-language description to control the object interactions, significantly reducing the human efforts for creating a large amount of interaction data. To facilitate the research of NLD simulation, we collect the Language-to-Interaction(L2I) benchmark dataset with 120,000 natural-language descriptions of object interactions in 6 common types of road topologies. Each description is associated with the programming code, which the graphic render can use to visually reconstruct the object interactions in the virtual scenes. As a methodology contribution, we design SimCopilot to translate the interaction descriptions to the renderable code. We use the L2I dataset to evaluate SimCopilot's abilities to control the object motions, generate complex interactions, and generalize interactions across road topologies. The L2I dataset and the evaluation results motivate the relevant research of the NLD simulation., Comment: 12 pages, 6 figures

Many methods of semantic image segmentation have borrowed the success of open compound domain adaptation. They minimize the style gap between the images of source and target domains, more easily predicting the accurate pseudo annotations for target domain's images that train segmentation network. The existing methods globally adapt the scene style of the images, whereas the object styles of different categories or instances are adapted improperly. This paper proposes the Object Style Compensation, where we construct the Object-Level Discrepancy Memory with multiple sets of discrepancy features. The discrepancy features in a set capture the style changes of the same category's object instances adapted from target to source domains. We learn the discrepancy features from the images of source and target domains, storing the discrepancy features in memory. With this memory, we select appropriate discrepancy features for compensating the style information of the object instances of various categories, adapting the object styles to a unified style of source domain. Our method enables a more accurate computation of the pseudo annotations for target domain's images, thus yielding state-of-the-art results on different datasets., Comment: Accepted by NeurlPS2023

Using synthesized images to boost the performance of perception models is a long-standing research challenge in computer vision. It becomes more eminent in visual-centric autonomous driving systems with multi-view cameras as some long-tail scenarios can never be collected. Guided by the BEV segmentation layouts, the existing generative networks seem to synthesize photo-realistic street-view images when evaluated solely on scene-level metrics. However, once zoom-in, they usually fail to produce accurate foreground and background details such as heading. To this end, we propose a two-stage generative method, dubbed BEVControl, that can generate accurate foreground and background contents. In contrast to segmentation-like input, it also supports sketch style input, which is more flexible for humans to edit. In addition, we propose a comprehensive multi-level evaluation protocol to fairly compare the quality of the generated scene, foreground object, and background geometry. Our extensive experiments show that our BEVControl surpasses the state-of-the-art method, BEVGen, by a significant margin, from 5.89 to 26.80 on foreground segmentation mIoU. In addition, we show that using images generated by BEVControl to train the downstream perception model, it achieves on average 1.29 improvement in NDS score., Comment: 13 pages, 8 figures

With the rapid development of high-resolution 3D vision applications, the traditional way of manipulating surface detail requires considerable memory and computing time. To address these problems, we introduce an efficient surface detail processing framework in 2D normal domain, which extracts new normal feature representations as the carrier of micro geometry structures that are illustrated both theoretically and empirically in this article. Compared with the existing state of the arts, we verify and demonstrate that the proposed normal-based representation has three important properties, including detail separability, detail transferability and detail idempotence. Finally, three new schemes are further designed for geometric surface detail processing applications, including geometric texture synthesis, geometry detail transfer, and 3D surface super-resolution. Theoretical analysis and experimental results on the latest benchmark dataset verify the effectiveness and versatility of our normal-based representation, which accepts 30 times of the input surface vertices but at the same time only takes 6.5% memory cost and 14.0% running time in comparison with existing competing algorithms., Comment: 16 pages, 22 figures

Semantic scene completion (SSC) requires an accurate understanding of the geometric and semantic relationships between the objects in the 3D scene for reasoning the occluded objects. The popular SSC methods voxelize the 3D objects, allowing the deep 3D convolutional network (3D CNN) to learn the object relationships from the complex scenes. However, the current networks lack the controllable kernels to model the object relationship across multiple views, where appropriate views provide the relevant information for suggesting the existence of the occluded objects. In this paper, we propose Cross-View Synthesis Transformer (CVSformer), which consists of Multi-View Feature Synthesis and Cross-View Transformer for learning cross-view object relationships. In the multi-view feature synthesis, we use a set of 3D convolutional kernels rotated differently to compute the multi-view features for each voxel. In the cross-view transformer, we employ the cross-view fusion to comprehensively learn the cross-view relationships, which form useful information for enhancing the features of individual views. We use the enhanced features to predict the geometric occupancies and semantic labels of all voxels. We evaluate CVSformer on public datasets, where CVSformer yields state-of-the-art results., Comment: 10 pages, 6 figures

Segment anything model (SAM) has presented impressive objectness identification capability with the idea of prompt learning and a new collected large-scale dataset. Given a prompt (e.g., points, bounding boxes, or masks) and an input image, SAM is able to generate valid segment masks for all objects indicated by the prompts, presenting high generalization across diverse scenarios and being a general method for zero-shot transfer to downstream vision tasks. Nevertheless, it remains unclear whether SAM may introduce errors in certain threatening scenarios. Clarifying this is of significant importance for applications that require robustness, such as autonomous vehicles. In this paper, we aim to study the testing-time robustness of SAM under adversarial scenarios and common corruptions. To this end, we first build a testing-time robustness evaluation benchmark for SAM by integrating existing public datasets. Second, we extend representative adversarial attacks against SAM and study the influence of different prompts on robustness. Third, we study the robustness of SAM under diverse corruption types by evaluating SAM on corrupted datasets with different prompts. With experiments conducted on SA-1B and KITTI datasets, we find that SAM exhibits remarkable robustness against various corruptions, except for blur-related corruption. Furthermore, SAM remains susceptible to adversarial attacks, particularly when subjected to PGD and BIM attacks. We think such a comprehensive study could highlight the importance of the robustness issues of SAM and trigger a series of new tasks for SAM as well as downstream vision tasks., Comment: 22 pages

Fully-supervised shadow removal methods achieve the best restoration qualities on public datasets but still generate some shadow remnants. One of the reasons is the lack of large-scale shadow & shadow-free image pairs. Unsupervised methods can alleviate the issue but their restoration qualities are much lower than those of fully-supervised methods. In this work, we find that pretraining shadow removal networks on the image inpainting dataset can reduce the shadow remnants significantly: a naive encoder-decoder network gets competitive restoration quality w.r.t. the state-of-the-art methods via only 10% shadow & shadow-free image pairs. After analyzing networks with/without inpainting pre-training via the information stored in the weight (IIW), we find that inpainting pretraining improves restoration quality in non-shadow regions and enhances the generalization ability of networks significantly. Additionally, shadow removal fine-tuning enables networks to fill in the details of shadow regions. Inspired by these observations we formulate shadow removal as an adaptive fusion task that takes advantage of both shadow removal and image inpainting. Specifically, we develop an adaptive fusion network consisting of two encoders, an adaptive fusion block, and a decoder. The two encoders are responsible for extracting the feature from the shadow image and the shadow-masked image respectively. The adaptive fusion block is responsible for combining these features in an adaptive manner. Finally, the decoder converts the adaptive fused features to the desired shadow-free result. The extensive experiments show that our method empowered with inpainting outperforms all state-of-the-art methods.

Scribble-supervised medical image segmentation tackles the limitation of sparse masks. Conventional approaches alternate between: labeling pseudo-masks and optimizing network parameters. However, such iterative two-stage paradigm is unwieldy and could be trapped in poor local optima since the networks undesirably regress to the erroneous pseudo-masks. To address these issues, we propose a non-iterative method where a stream of varying (pacing) pseudo-masks teach a network via consistency training, named PacingPseudo. Our motivation lies first in a non-iterative process. Interestingly, it can be achieved gracefully by a siamese architecture, wherein a stream of pseudo-masks naturally assimilate a stream of predicted masks during training. Second, we make the consistency training effective with two necessary designs: (i) entropy regularization to obtain high-confidence pseudo-masks for effective teaching; and (ii) distorted augmentations to create discrepancy between the pseudo-mask and predicted-mask streams for consistency regularization. Third, we devise a new memory bank mechanism that provides an extra source of ensemble features to complement scarce labeled pixels. The efficacy of the proposed PacingPseudo is validated on three public medical image datasets, including the segmentation tasks of abdominal multi-organs, cardiac structures, and myocardium. Extensive experiments demonstrate our PacingPseudo improves the baseline by large margins and consistently outcompetes several previous methods. In some cases, our PacingPseudo achieves comparable performance with its fully-supervised counterparts, showing the feasibility of our method for the challenging scribble-supervised segmentation applications. The code and scribble annotations will be publicly available., Comment: 12 pages, 8 figures

Although achieving significant progress, existing deep generative inpainting methods are far from real-world applications due to the low generalization across different scenes. As a result, the generated images usually contain artifacts or the filled pixels differ greatly from the ground truth. Image-level predictive filtering is a widely used image restoration technique, predicting suitable kernels adaptively according to different input scenes. Inspired by this inherent advantage, we explore the possibility of addressing image inpainting as a filtering task. To this end, we first study the advantages and challenges of image-level predictive filtering for image inpainting: the method can preserve local structures and avoid artifacts but fails to fill large missing areas. Then, we propose semantic filtering by conducting filtering on the deep feature level, which fills the missing semantic information but fails to recover the details. To address the issues while adopting the respective advantages, we propose a novel filtering technique, i.e., Multilevel Interactive Siamese Filtering (MISF), which contains two branches: kernel prediction branch (KPB) and semantic & image filtering branch (SIFB). These two branches are interactively linked: SIFB provides multi-level features for KPB while KPB predicts dynamic kernels for SIFB. As a result, the final method takes the advantage of effective semantic & image-level filling for high-fidelity inpainting. We validate our method on three challenging datasets, i.e., Dunhuang, Places2, and CelebA. Our method outperforms state-of-the-art baselines on four metrics, i.e., L1, PSNR, SSIM, and LPIPS. Please try the released code and model at https://github.com/tsingqguo/misf., Comment: Accepted by CVPR 2022

Deraining is a significant and fundamental computer vision task, aiming to remove the rain streaks and accumulations in an image or video captured under a rainy day. Existing deraining methods usually make heuristic assumptions of the rain model, which compels them to employ complex optimization or iterative refinement for high recovery quality. This, however, leads to time-consuming methods and affects the effectiveness for addressing rain patterns deviated from from the assumptions. In this paper, we propose a simple yet efficient deraining method by formulating deraining as a predictive filtering problem without complex rain model assumptions. Specifically, we identify spatially-variant predictive filtering (SPFilt) that adaptively predicts proper kernels via a deep network to filter different individual pixels. Since the filtering can be implemented via well-accelerated convolution, our method can be significantly efficient. We further propose the EfDeRain+ that contains three main contributions to address residual rain traces, multi-scale, and diverse rain patterns without harming the efficiency. First, we propose the uncertainty-aware cascaded predictive filtering (UC-PFilt) that can identify the difficulties of reconstructing clean pixels via predicted kernels and remove the residual rain traces effectively. Second, we design the weight-sharing multi-scale dilated filtering (WS-MS-DFilt) to handle multi-scale rain streaks without harming the efficiency. Third, to eliminate the gap across diverse rain patterns, we propose a novel data augmentation method (i.e., RainMix) to train our deep models. By combining all contributions with sophisticated analysis on different variants, our final method outperforms baseline methods on four single-image deraining datasets and one video deraining dataset in terms of both recovery quality and speed., Comment: 14 pages, 10 figures, 10 tables. This is the extention of our conference version https://github.com/tsingqguo/efficientderain

Automatic segmentation of abdominal organs in computed tomography (CT) images can support radiation therapy and image-guided surgery workflows. Developing of such automatic solutions remains challenging mainly owing to complex organ interactions and blurry boundaries in CT images. To address these issues, we focus on effective spatial context modeling and explicit edge segmentation priors. Accordingly, we propose a 3D network with four main components trained end-to-end including shared encoder, edge detector, decoder with edge skip-connections (ESCs) and recurrent feature propagation head (RFP-Head). To capture wide-range spatial dependencies, the RFP-Head propagates and harvests local features through directed acyclic graphs (DAGs) formulated with recurrent connections in an efficient slice-wise manner, with regard to spatial arrangement of image units. To leverage edge information, the edge detector learns edge prior knowledge specifically tuned for semantic segmentation by exploiting intermediate features from the encoder with the edge supervision. The ESCs then aggregate the edge knowledge with multi-level decoder features to learn a hierarchy of discriminative features explicitly modeling complementarity between organs' interiors and edges for segmentation. We conduct extensive experiments on two challenging abdominal CT datasets with eight annotated organs. Experimental results show that the proposed network outperforms several state-of-the-art models, especially for the segmentation of small and complicated structures (gallbladder, esophagus, stomach, pancreas and duodenum). The code will be publicly available.

3D point cloud completion is very challenging because it heavily relies on the accurate understanding of the complex 3D shapes (e.g., high-curvature, concave/convex, and hollowed-out 3D shapes) and the unknown & diverse patterns of the partially available point clouds. In this paper, we propose a novel solution,i.e., Point-block Carving (PC), for completing the complex 3D point cloud completion. Given the partial point cloud as the guidance, we carve a3D block that contains the uniformly distributed 3D points, yielding the entire point cloud. To achieve PC, we propose a new network architecture, i.e., CarveNet. This network conducts the exclusive convolution on each point of the block, where the convolutional kernels are trained on the 3D shape data. CarveNet determines which point should be carved, for effectively recovering the details of the complete shapes. Furthermore, we propose a sensor-aware method for data augmentation,i.e., SensorAug, for training CarveNet on richer patterns of partial point clouds, thus enhancing the completion power of the network. The extensive evaluations on the ShapeNet and KITTI datasets demonstrate the generality of our approach on the partial point clouds with diverse patterns. On these datasets, CarveNet successfully outperforms the state-of-the-art methods., Comment: 10 pages and 10 figures

Orbital angular momentum (OAM) based structured light beams provide an additional degree of freedom for practical applications ranging from optical communication to laser-based material processing. Many techniques exist for generating such beams within laser sources and these primarily rely upon the use of specially designed optical components that limit laser power scaling and ready tunability of the topological charge and polarization of the output OAM beams. Here we show that some of these limitations can be overcome by employing a computer controlled reflective phase-only spatial light modulator (SLM) to adaptively tailor the input (and subsequent output) beam wavefront and polarization in a few-mode fibre amplifier. In this way modal-coupling induced beam distortion within the fibre amplifier can be mitigated and we are able to generate at will any desired supported spatial mode guided in the fibre, including conventional LP modes, scalar OAM modes and cylindrical vector modes, at average powers >10 W and with a peak power of >11 kW. Our results pave the way to the realization of practical high-power structured laser sources with tunable chirality and polarization.