Incident management remains a critical yet challenging task for large-scale cloud services. Most cloud service providers abstract troubleshooting into predefined workflows for different incidents, offering step-by-step guidance. However, manually crafting workflows is resource-consuming and knowledge-intensive, hindering large-scale deployment. Most automated techniques for workflow orchestration rely on large language models (LLMs) to handle complex tasks but overlook key aspects of troubleshooting, including complex expertise, domain requirements, and the reliability of AI feedback. These limitations undermine workflow quality. Therefore, we propose FlowXpert, a novel framework for troubleshooting workflow orchestration. Leveraging LLMs, it first builds a knowledge base centered on incident-aware nodes to precisely depict expertise. Then, fed into AI feedback and synthetic preference data, reinforcement learning is applied to refine the workflow generator and evaluator. To assess troubleshooting workflows, we introduce OpsFlowBench based on Huawei Cloud's datacenter switch operation documents. Benchmark tests under the tailored STEPScore metric validate its effectiveness. Furthermore, during a 10-week deployment in Huawei Cloud's datacenter network, FlowXpert provided valuable support to both on-call engineers and AI executors, as evidenced by empirical data and case study.

The application of deep neural networks (DNNs) on 360-degree images has achieved remarkable progress in the recent years. However, DNNs have been demonstrated to be vulnerable to well-crafted adversarial examples, which may trigger severe safety problems in the real-world applications based on 360-degree images. In this paper, we propose an adversarial attack targeting spherical images, called 360-attactk, that transfers adversarial perturbations from perspective-view (PV) images to a final adversarial spherical image. Given a target spherical image, we first represent it with a set of planar PV images, and then perform 2D attacks on them to obtain adversarial PV images. Considering the issue of the projective distortion between spherical and PV images, we propose a distortion-aware attack to reduce the negative impact of distortion on attack. Moreover, to reconstruct the final adversarial spherical image with high aggressiveness, we calculate the spherical saliency map with a novel spherical spectrum method and next propose a saliency-aware fusion strategy that merges multiple inverse perspective projections for the same position on the spherical image. Extensive experimental results show that 360-attack is effective for disturbing spherical images in the black-box setting. Our attack also proves the presence of adversarial transferability from Z2 to SO(3) groups.

Depth estimation for panorama is a key part of 3D scene understanding, and adopting discriminative models is the most common solution. However, due to the rectangular convolution kernel, these existing learning methods cannot efficiently extract the distorted features in panoramas. To this end, we propose OmniVAE, a generative model based on Conditional Variational Auto-Encoder (CVAE) and von Mises-Fisher (vMF) distribution, to strengthen the exclusive generative ability for spherical signals by mapping panoramas to hypersphere space. Further, to alleviate the side effects of manifold-mismatching caused by non-planar distribution, we put forward the Atypical Receptive Field (ARF) module to slightly shift the receptive field of the network and even take the distribution difference into account in the reconstruction loss. The quantitative and qualitative evaluations are performed on real-world and synthetic datasets, and the results show that OmniVAE outperforms the state-of-the-art methods.

Panoramic images are usually viewed through Head Mounted Displays (HMDs), which renders only a narrow field of view from the raw panoramic image. This distinctive viewing feature has largely been ignored when inpainting panoramic images. To address this issue, we propose a viewport-oriented generative adversarial panoramic image inpainting network in this paper. For capturing the distorted features accurately in the generating process of equirectangular projection (ERP) panoramic image, a latitude-adaptive feature fusion module is devised to aggregate the latitude-level features in ERP image and less-distorted patch-level viewport-domain features. Furthermore, a novel cross-domain discriminator is proposed to force the inpainting network to generate more plausible results in viewports. Extensive experiments show that our model achieves better performance compared to the baseline methods, especially in the viewport images.

Object detection methods for perspective images have proven increasingly efficient, but the techniques for equirectangular projection (ERP) panoramas from inherently spherical imaging cannot still achieve satisfactory performance. Due to the various degrees of distortion at different pixel locations, current algorithms cannot adapt to the changes in shape and contour caused by stretching, which results in performance degradation when migrating them from perspective images to spherical ones. In this paper, we improve the network for panorama object detection and introduce the cube-domain information with discontinuity but low distortion to correct the panorama features. Unlike previous works, we consider the impact of semantic discontinuity from all tangent planes instead of overlaying features when needed. Considering the six facets as unified, i.e., six-to-one for extraction, the proposed Facet-Link module enhances the long-range sensing capability at the facet level in the frequency domain. Moreover, the position alignment packs different facets, i.e., six-to-one for calibration, to preserve more global signals during the correction stage, which establishes semantic pathways for feature interactions between panorama and cubemap in the two dimensions, facet-facet and cube-pano, respectively. Extensive experiments on synthetic and real-world datasets verify the effectiveness and robustness of our proposed method.

As one type of omnidirectional projection, fisheye images have been widely used in automatic driving and visual surveillance. However, they cannot be processed well by the traditional algorithms designed for the planar rectilinear images since they usually suffer from severe geometric distortion during image formation. In this paper, the conventional face detection algorithm is enhanced to fit the fisheye images via combining with the spherical convolution block by learning rotation-invariant features from the spherical domain. The learned features from both planar and spherical domains are subsequently mixed by the spatial attention mechanism. Consequently, the whole network can automatically learn the distorted features directly from different positions on the target image. Experimental results verify that our network can detect distorted faces on fisheye images effectively and maintain the performance on traditional planar images.