Search

Self-supervised learning usually uses a large amount of unlabeled data to pre-train an encoder which can be used as a general-purpose feature extractor, such that downstream users only need to perform fine-tuning operations to enjoy the benefit of "large model". Despite this promising prospect, the security of pre-trained encoder has not been thoroughly investigated yet, especially when the pre-trained encoder is publicly available for commercial use. In this paper, we propose AdvEncoder, the first framework for generating downstream-agnostic universal adversarial examples based on the pre-trained encoder. AdvEncoder aims to construct a universal adversarial perturbation or patch for a set of natural images that can fool all the downstream tasks inheriting the victim pre-trained encoder. Unlike traditional adversarial example works, the pre-trained encoder only outputs feature vectors rather than classification labels. Therefore, we first exploit the high frequency component information of the image to guide the generation of adversarial examples. Then we design a generative attack framework to construct adversarial perturbations/patches by learning the distribution of the attack surrogate dataset to improve their attack success rates and transferability. Our results show that an attacker can successfully attack downstream tasks without knowing either the pre-training dataset or the downstream dataset. We also tailor four defenses for pre-trained encoders, the results of which further prove the attack ability of AdvEncoder., Comment: This paper has been accepted by the International Conference on Computer Vision (ICCV '23, October 2--6, 2023, Paris, France)

This paper addresses the problem of cross-modal object tracking from RGB videos and event data. Rather than constructing a complex cross-modal fusion network, we explore the great potential of a pre-trained vision Transformer (ViT). Particularly, we delicately investigate plug-and-play training augmentations that encourage the ViT to bridge the vast distribution gap between the two modalities, enabling comprehensive cross-modal information interaction and thus enhancing its ability. Specifically, we propose a mask modeling strategy that randomly masks a specific modality of some tokens to enforce the interaction between tokens from different modalities interacting proactively. To mitigate network oscillations resulting from the masking strategy and further amplify its positive effect, we then theoretically propose an orthogonal high-rank loss to regularize the attention matrix. Extensive experiments demonstrate that our plug-and-play training augmentation techniques can significantly boost state-of-the-art one-stream and twostream trackers to a large extent in terms of both tracking precision and success rate. Our new perspective and findings will potentially bring insights to the field of leveraging powerful pre-trained ViTs to model cross-modal data. The code will be publicly available.

Quantifying the dissimilarity between two unstructured 3D point clouds is a challenging task, with existing metrics often relying on measuring the distance between corresponding points that can be either inefficient or ineffective. In this paper, we propose a novel distance metric called Calibrated Local Geometry Distance (CLGD), which computes the difference between the underlying 3D surfaces calibrated and induced by a set of reference points. By associating each reference point with two given point clouds through computing its directional distances to them, the difference in directional distances of an identical reference point characterizes the geometric difference between a typical local region of the two point clouds. Finally, CLGD is obtained by averaging the directional distance differences of all reference points. We evaluate CLGD on various optimization and unsupervised learning-based tasks, including shape reconstruction, rigid registration, scene flow estimation, and feature representation. Extensive experiments show that CLGD achieves significantly higher accuracy under all tasks in a memory and computationally efficient manner, compared with existing metrics. As a generic metric, CLGD has the potential to advance 3D point cloud modeling. The source code is publicly available at https://github.com/rsy6318/CLGD.

The challenge of dynamic view synthesis from dynamic monocular videos, i.e., synthesizing novel views for free viewpoints given a monocular video of a dynamic scene captured by a moving camera, mainly lies in accurately modeling the dynamic objects of a scene using limited 2D frames, each with a varying timestamp and viewpoint. Existing methods usually require pre-processed 2D optical flow and depth maps by additional methods to supervise the network, making them suffer from the inaccuracy of the pre-processed supervision and the ambiguity when lifting the 2D information to 3D. In this paper, we tackle this challenge in an unsupervised fashion. Specifically, we decouple the motion of the dynamic objects into object motion and camera motion, respectively regularized by proposed unsupervised surface consistency and patch-based multi-view constraints. The former enforces the 3D geometric surfaces of moving objects to be consistent over time, while the latter regularizes their appearances to be consistent across different viewpoints. Such a fine-grained motion formulation can alleviate the learning difficulty for the network, thus enabling it to produce not only novel views with higher quality but also more accurate scene flows and depth than existing methods requiring extra supervision. We will make the code publicly available.

In recent years, point clouds have become increasingly popular for representing three-dimensional (3D) visual objects and scenes. To efficiently store and transmit point clouds, compression methods have been developed, but they often result in a degradation of quality. To reduce color distortion in point clouds, we propose a graph-based quality enhancement network (GQE-Net) that uses geometry information as an auxiliary input and graph convolution blocks to extract local features efficiently. Specifically, we use a parallel-serial graph attention module with a multi-head graph attention mechanism to focus on important points or features and help them fuse together. Additionally, we design a feature refinement module that takes into account the normals and geometry distance between points. To work within the limitations of GPU memory capacity, the distorted point cloud is divided into overlap-allowed 3D patches, which are sent to GQE-Net for quality enhancement. To account for differences in data distribution among different color omponents, three models are trained for the three color components. Experimental results show that our method achieves state-of-the-art performance. For example, when implementing GQE-Net on the recent G-PCC coding standard test model, 0.43 dB, 0.25 dB, and 0.36 dB Bjontegaard delta (BD)-peak-signal-to-noise ratio (PSNR), corresponding to 14.0%, 9.3%, and 14.5% BD-rate savings can be achieved on dense point clouds for the Y, Cb, and Cr components, respectively., 13 pages, 11 figures, submitted to IEEE TIP

Point cloud sampling can reduce storage requirements and computation costs for various vision tasks. Traditional sampling methods, such as farthest point sampling, are not geared towards downstream tasks and may fail on such tasks. In this paper, we propose a cascade attention-based sampling network (CAS-Net), which is end-to-end trainable. Specifically, we propose an attention-based sampling module (ASM) to capture the semantic features and preserve the geometry of the original point cloud. Experimental results on the ModelNet40 dataset show that CAS-Net outperforms state-of-the-art methods in a sampling-based point cloud classification task, while preserving the geometric structure of the sampled point cloud.

Recent works have revealed the superiority of feature-level fusion for cross-modal 3D object detection, where fine-grained feature propagation from 2D image pixels to 3D LiDAR points has been widely adopted for performance improvement. Still, the potential of heterogeneous feature propagation between 2D and 3D domains has not been fully explored. In this paper, in contrast to existing pixel-to-point feature propagation, we investigate an opposite point-to-pixel direction, allowing point-wise features to flow inversely into the 2D image branch. Thus, when jointly optimizing the 2D and 3D streams, the gradients back-propagated from the 2D image branch can boost the representation ability of the 3D backbone network working on LiDAR point clouds. Then, combining pixel-to-point and point-to-pixel information flow mechanisms, we construct an bidirectional feature propagation framework, dubbed BiProDet. In addition to the architectural design, we also propose normalized local coordinates map estimation, a new 2D auxiliary task for the training of the 2D image branch, which facilitates learning local spatial-aware features from the image modality and implicitly enhances the overall 3D detection performance. Extensive experiments and ablation studies validate the effectiveness of our method. Notably, we rank $\mathbf{1^{\mathrm{st}}}$ on the highly competitive KITTI benchmark on the cyclist class by the time of submission. The source code is available at https://github.com/Eaphan/BiProDet.

In this paper, we study the problem of embedding the high-dimensional spatio-spectral information of hyperspectral (HS) images efficiently and effectively, oriented by feature diversity. To be specific, based on the theoretical formulation that feature diversity is correlated with the rank of the unfolded kernel matrix, we rectify 3D convolution by modifying its topology to boost the rank upper-bound, yielding a rank-enhanced spatial-spectral symmetrical convolution set (ReS$^3$-ConvSet), which is able to not only learn diverse and powerful feature representations but also save network parameters. In addition, we also propose a novel diversity-aware regularization (DA-Reg) term, which acts directly on the feature maps to maximize the independence among elements. To demonstrate the superiority of the proposed ReS$^3$-ConvSet and DA-Reg, we apply them to various HS image processing and analysis tasks, including denoising, spatial super-resolution, and classification. Extensive experiments demonstrate that the proposed approaches outperform state-of-the-art methods to a significant extent both quantitatively and qualitatively. The code is publicly available at \url{https://github.com/jinnh/ReSSS-ConvSet}., 15 pages, 12 figures. arXiv admin note: substantial text overlap with arXiv:2207.04266

The past few years have witnessed the prevalence of self-supervised representation learning within the language and 2D vision communities. However, such advancements have not been fully migrated to the 3D point cloud learning community. Different from existing pre-training paradigms for 3D point clouds falling into the scope of generative modeling or contrastive learning, this paper proposes a translative pre-training framework, namely PointVST, driven by a novel self-supervised pretext task of cross-modal translation from 3D point clouds to their corresponding diverse forms of 2D rendered images. More specifically, we start by deducing view-conditioned point-wise embeddings via the insertion of a viewpoint indicator and then adaptively aggregate a view-specific global codeword fed into the subsequent 2D convolutional translation heads for image generation. Extensive experiments on various downstream tasks of 3D shape analysis and scene understanding demonstrate that PointVST shows consistent and prominent performance superiority over current state-of-the-art methods. Our code will be made publicly available.

Deep learning-based 3D object detectors have made significant progress in recent years and have been deployed in a wide range of applications. It is crucial to understand the robustness of detectors against adversarial attacks when employing detectors in security-critical applications. In this paper, we make the first attempt to conduct a thorough evaluation and analysis of the robustness of 3D detectors under adversarial attacks. Specifically, we first extend three kinds of adversarial attacks to the 3D object detection task to benchmark the robustness of state-of-the-art 3D object detectors against attacks on KITTI and Waymo datasets, subsequently followed by the analysis of the relationship between robustness and properties of detectors. Then, we explore the transferability of cross-model, cross-task, and cross-data attacks. We finally conduct comprehensive experiments of defense for 3D detectors, demonstrating that simple transformations like flipping are of little help in improving robustness when the strategy of transformation imposed on input point cloud data is exposed to attackers. Our findings will facilitate investigations in understanding and defending the adversarial attacks against 3D object detectors to advance this field.

Existing graph clustering networks heavily rely on a predefined graph and may fail if the initial graph is of low quality. To tackle this issue, we propose a novel graph augmentation clustering network capable of adaptively enhancing the initial graph to achieve better clustering performance. Specifically, we first integrate the node attribute and topology structure information to learn the latent feature representation. Then, we explore the local geometric structure information on the embedding space to construct an adjacency graph and subsequently develop an adaptive graph augmentation architecture to fuse that graph with the initial one dynamically. Finally, we minimize the Jeffreys divergence between multiple derived distributions to conduct network training in an unsupervised fashion. Extensive experiments on six commonly used benchmark datasets demonstrate that the proposed method consistently outperforms several state-of-the-art approaches. In particular, our method improves the ARI by more than 9.39\% over the best baseline on DBLP. The source codes and data have been submitted to the appendix.

As two fundamental representation modalities of 3D objects, 3D point clouds and multi-view 2D images record shape information from different domains of geometric structures and visual appearances. In the current deep learning era, remarkable progress in processing such two data modalities has been achieved through respectively customizing compatible 3D and 2D network architectures. However, unlike multi-view image-based 2D visual modeling paradigms, which have shown leading performance in several common 3D shape recognition benchmarks, point cloud-based 3D geometric modeling paradigms are still highly limited by insufficient learning capacity, due to the difficulty of extracting discriminative features from irregular geometric signals. In this paper, we explore the possibility of boosting deep 3D point cloud encoders by transferring visual knowledge extracted from deep 2D image encoders under a standard teacher-student distillation workflow. Generally, we propose PointMCD, a unified multi-view cross-modal distillation architecture, including a pretrained deep image encoder as the teacher and a deep point encoder as the student. To perform heterogeneous feature alignment between 2D visual and 3D geometric domains, we further investigate visibility-aware feature projection (VAFP), by which point-wise embeddings are reasonably aggregated into view-specific geometric descriptors. By pair-wisely aligning multi-view visual and geometric descriptors, we can obtain more powerful deep point encoders without exhausting and complicated network modification. Experiments on 3D shape classification, part segmentation, and unsupervised learning strongly validate the effectiveness of our method. The code and data will be publicly available at https://github.com/keeganhk/PointMCD., Comment: Accepted to TMM

The inherent ambiguity in ground-truth annotations of 3D bounding boxes, caused by occlusions, signal missing, or manual annotation errors, can confuse deep 3D object detectors during training, thus deteriorating detection accuracy. However, existing methods overlook such issues to some extent and treat the labels as deterministic. In this paper, we formulate the label uncertainty problem as the diversity of potentially plausible bounding boxes of objects. Then, we propose GLENet, a generative framework adapted from conditional variational autoencoders, to model the one-to-many relationship between a typical 3D object and its potential ground-truth bounding boxes with latent variables. The label uncertainty generated by GLENet is a plug-and-play module and can be conveniently integrated into existing deep 3D detectors to build probabilistic detectors and supervise the learning of the localization uncertainty. Besides, we propose an uncertainty-aware quality estimator architecture in probabilistic detectors to guide the training of the IoU-branch with predicted localization uncertainty. We incorporate the proposed methods into various popular base 3D detectors and demonstrate significant and consistent performance gains on both KITTI and Waymo benchmark datasets. Especially, the proposed GLENet-VR outperforms all published LiDAR-based approaches by a large margin and achieves the top rank among single-modal methods on the challenging KITTI test set. The source code and pre-trained models are publicly available at \url{https://github.com/Eaphan/GLENet}.

Depth estimation is one of the most essential problems for light field applications. In EPI-based methods, the slope computation usually suffers low accuracy due to the discretization error and low angular resolution. In addition, recent methods work well in most regions but often struggle with blurry edges over occluded regions and ambiguity over texture-less regions. To address these challenging issues, we first propose the stitched-EPI and half-stitched-EPI algorithms for non-occluded and occluded regions, respectively. The algorithms improve slope computation by shifting and concatenating lines in different EPIs but related to the same point in 3D scene, while the half-stitched-EPI only uses non-occluded part of lines. Combined with the joint photo-consistency cost proposed by us, the more accurate and robust depth map can be obtained in both occluded and non-occluded regions. Furthermore, to improve the depth estimation in texture-less regions, we propose a depth propagation strategy that determines their depth from the edge to interior, from accurate regions to coarse regions. Experimental and ablation results demonstrate that the proposed method achieves accurate and robust depth maps in all regions effectively., Comment: 15 pages

This paper explores the problem of reconstructing high-resolution light field (LF) images from hybrid lenses, including a high-resolution camera surrounded by multiple low-resolution cameras. The performance of existing methods is still limited, as they produce either blurry results on plain textured areas or distortions around depth discontinuous boundaries. To tackle this challenge, we propose a novel end-to-end learning-based approach, which can comprehensively utilize the specific characteristics of the input from two complementary and parallel perspectives. Specifically, one module regresses a spatially consistent intermediate estimation by learning a deep multidimensional and cross-domain feature representation, while the other module warps another intermediate estimation, which maintains the high-frequency textures, by propagating the information of the high-resolution view. We finally leverage the advantages of the two intermediate estimations adaptively via the learned attention maps, leading to the final high-resolution LF image with satisfactory results on both plain textured areas and depth discontinuous boundaries. Besides, to promote the effectiveness of our method trained with simulated hybrid data on real hybrid data captured by a hybrid LF imaging system, we carefully design the network architecture and the training strategy. Extensive experiments on both real and simulated hybrid data demonstrate the significant superiority of our approach over state-of-the-art ones. To the best of our knowledge, this is the first end-to-end deep learning method for LF reconstruction from a real hybrid input. We believe our framework could potentially decrease the cost of high-resolution LF data acquisition and benefit LF data storage and transmission., Accepted by IEEE TPAMI. arXiv admin note: text overlap with arXiv:1907.09640