This paper looks at this intriguing question: are single images with their details lost during deraining, reversible to artifact-free status? We propose an end-to-end detail-recovery image deraining network (termed a DRDNet) solve the problem. Unlike existing approaches that attempt meet conflicting goal of simultaneously and preserving in unified framework, we view rain removal detail recovery as two seperate tasks, so each part could specialize rather than trade-off between goals....

Convolution on 3D point clouds that generalized from 2D grid-like domains is widely researched yet far perfect. The standard convolution characterises feature correspondences indistinguishably among points, presenting an intrinsic limitation of poor distinctive learning. In this paper, we propose Adaptive Graph (AdaptConv) which generates adaptive kernels for points according to their dynamically learned features. Compared with using a fixed/isotropic kernel, AdaptConv improves the...

Can you find me? By simulating how humans to discover the so-called 'perfectly'-camouflaged object, we present a novel boundary-guided separated attention network (call BSA-Net). Beyond existing camouflaged object detection (COD) wisdom, BSA-Net utilizes two-stream modules highlight separator (or say object's boundary) between an image's background and foreground: reverse stream helps erase interior focus on background, while normal recovers thus pay more foreground; both streams are...

Low-light image enhancement (LLE) remains challenging due to the unfavorable prevailing low-contrast and weak-visibility problems of single RGB images. In this paper, we respond intriguing learning-related question -- if leveraging both accessible unpaired over/underexposed images high-level semantic guidance, can improve performance cutting-edge LLE models? Here, propose an effective semantically contrastive learning paradigm for (namely SCL-LLE). Beyond existing wisdom, it casts task as...

While the wisdom of training an image dehazing model on synthetic hazy data can alleviate difficulty collecting real-world hazy/clean pairs, it brings well-known domain shift problem. From a different yet new perspective, this paper explores contrastive learning with adversarial effort to leverage unpaired and clean images, thus alleviating problem enhancing network's generalization ability in scenarios. We propose effective unsupervised paradigm for dehazing, dubbed UCL-Dehaze. Unpaired...

Image dehazing is a common operation in autonomous driving, traffic monitoring and surveillance. Learning-based image has achieved excellent performance recently. However, it nearly impossible to capture pairs of hazy/clean images from the real world train an network. Most existing models that are learnt synthetically generated hazy generalize poorly on real-world scenarios due obvious domain shift. To deal with this unpaired problem arisen by images, we present Cycle Spectral Normalized...

Convolution on 3D point clouds is widely researched yet far from perfect in geometric deep learning. The traditional wisdom of convolution characterises feature correspondences indistinguishably among points, arising an intrinsic limitation poor distinctive In this article, we propose Adaptive Graph (AGConv) for wide applications cloud analysis. AGConv generates adaptive kernels points according to their dynamically learned features. Compared with the solution using fixed/isotropic kernels,...

Problems such as equipment defects or limited view-points will lead the captured point clouds to be incomplete. Therefore, recovering complete from partial ones plays an vital role in many practical tasks, and one of keys lies prediction missing part. In this paper, we propose a novel cloud completion approach namely ProxyFormer that divides into existing (input) (to predicted) parts each part communicates information through its proxies. Specifically, fuse proxy via feature position...

Snow is one of the toughest adverse weather conditions for object detection (OD). Currently, not only there a lack snowy OD datasets to train cutting-edge detectors, but also these detectors have difficulties learning latent information beneficial in snow. To alleviate two above problems, we first establish real-world dataset, named RSOD. Besides, develop an unsupervised training strategy with distinctive activation function, called <inline-formula...

How will you repair a physical object with some missings? You may imagine its original shape from previously captured images, recover overall (global) but coarse first, and then refine local details. We are motivated to imitate the procedure address point cloud completion. To this end, we propose cross-modal shape-transfer dual-refinement network (termed CSDN), coarse-to-fine paradigm images of full-cycle participation, for quality CSDN mainly consists "shape fusion" "dual-refinement"...

Haze severely degrades the visibility of scene objects and deteriorates performance autonomous driving, traffic monitoring, other vision-based intelligent transportation systems. As a potential remedy, we propose novel unified Transformer with semantically contrastive learning for image dehazing, dubbed USCFormer. USCFormer has three key contributions. First, absorbs respective strengths CNN by incorporating them into format. Thus, it allows simultaneous capture global-local dependency...

Most mesh denoising techniques utilize only either the facet normal field or vertex of a surface. The two fields, though contain some redundant geometry information same model, can provide additional that other lacks. Thus, considering one is likely to overlook geometric features. In this paper, we take advantage piecewise consistent property fields and propose an effective framework in which they are filtered integrated using novel method guide process. Our key observation that, decomposing...

Mesh denoising is imperative for improving imperfect surfaces acquired by scanning devices. The main challenge to faithfully retain geometric features and avoid introducing additional artifacts when removing noise. Unlike the existing mesh techniques that focus only on either first-order or high-order differential properties, our approach exploits synergy facet normals quadric are integrated recover a piecewise smooth surface. In specific, we vote surface normal tensors from robust...

Bas-relief is characterized by its unique presentation of intrinsic shape properties and/or detailed appearance using materials raised up in different degrees above a background. However, many bas-relief modeling methods could not manipulate scene details well. We propose simple and effective solution for two kinds (i.e., structure-preserving detail-preserving) which from the prior tone mapping alike methods. Our idea originates an observation on typical 3D models, are decomposed into...

Mesh denoising is a classical, yet not well-solved problem in digital geometry processing. The challenge arises from noise removal with the minimal disturbance of surface intrinsic properties (e.g., sharp features and shallow details). We propose new patch normal co-filter (PcFilter) for mesh denoising. It inspired by statistics which show that patches similar exist on underlying noisy mesh. model PcFilter as low-rank matrix recovery similar-patch collaboration, aiming at removing different...

Point cloud is the primary source from 3D scanners and depth cameras. It usually contains more raw geometric features, as well higher levels of noise than reconstructed mesh. Although many mesh denoising methods have proven to be effective in removal, they hardly work on noisy point clouds. We propose a new multi-patch collaborative method for denoising, which solved low-rank matrix recovery problem. Unlike traditional single-patch based approaches, our approach inspired by statistics...

Object detection has made tremendous strides in computer vision. Small object with appearance degradation is a prominent challenge, especially for aerial observations. To collect sufficient positive/negative samples heuristic training, most detectors preset region anchors order to calculate intersection-over-union (IoU) against the ground-truth data. In this case, small objects are frequently abandoned or mislabeled. article, we present an effective dynamic enhancement anchor network...

Hough voting, as has been demonstrated in VoteNet, is effective for 3D object detection, where voting a key step. In this paper, we propose novel VoteNet-based detector with vote enhancement to improve the detection accuracy cluttered indoor scenes. It addresses limitations of current schemes, i.e., votes from neighboring objects and background have significant negative impacts. Before replace classic MLP proposed Attentive (AMLP) backbone network get better feature description seed points....

Camouflaged objects share very similar colors but have different semantics with the surroundings. Cognitive scientists observe that both global contour (i.e., boundary) and local pattern texture) of camouflaged are key cues to help humans find them successfully. Inspired by cognitive scientist's observation, we propose a novel boundary-and-texture enhancement network (FindNet) for object detection (COD) from single images. Different most existing COD methods, FindNet embeds information into...

Object detection is one of the most fundamental yet challenging research topics in domain computer vision. Recently, study on this topic aerial images has made tremendous progress. However, complex background and worse imaging quality are obvious problems object detection. Most state-of-the-art approaches tend to develop elaborate attention mechanisms for space-time feature calibrations with arduous computational complexity, while surprisingly ignoring importance channel-wise. In work, we...