Surface reconstruction for micro-samples with large discontinuities using digital holography is a challenge. To overcome this problem, multi-incidence digital holographic profilometry (MIDHP) has been proposed. MIDHP relies on the numerical generation of the longitudinal scanning function (LSF) for reconstructing the topography of the sample with large depth and high axial resolution. Nevertheless, the method is unable to reconstruct surfaces with large gradients due to the need of: (i) high precision focusing that manual adjustment cannot fulfill and (ii) preserving the functionality of the LSF that requires capturing and processing many digital holograms. In this work, we propose a novel MIDHP method to solve these limitations. First, an autofocusing algorithm based on the comparison of shapes obtained by the LSF and the thin tilted element approximation is proposed. It is proven that this autofocusing algorithm is capable to deliver in-focus plane localization with submicron resolution. Second, we propose that wavefield summation for the generation of the LSF is carried out in Fourier space. It is shown that this scheme enables a significant reduction of arithmetic operations and can minimize the number of Fourier transforms needed. Hence, a fast generation of the LSF is possible without compromising its accuracy. The functionality of MIDHP for measuring surfaces with large gradients is supported by numerical and experimental results.

A panoptic segmentation network to predict masks and classes for things and stuff in images is proposed in this work. Recently, panoptic segmentation has been advanced through the combination of the query-based learning and end-to-end learning approaches. Current research focuses on learning queries without distinguishing between thing and stuff classes. We present decoupling query learning to generate effective thing and stuff queries for panoptic segmentation. For this purpose, we adopt different workflows for thing and stuff queries. We design center-guided query selection for thing queries, which focuses on the center regions of individual instances in images, while we set stuff queries as randomly initialized embeddings. Also, we apply a decoupling mask to the self-attention of query features to prevent interactions between things and stuff. In the query selection process, we generate a center heatmap that guides thing query selection. Experimental results demonstrate that the proposed panoptic segmentation network outperforms the state of the art on two panoptic segmentation datasets.

In recent years, there has been a noticeable increase in the inclination towards digitizing our surroundings, encompassing various domains such as virtual reality, cultural heritage conservation, and architectural representation. The computation of high-resolution three-dimensional (3D) colored point clouds and meshes holds significant importance for such applications. However, traditional structure-from-motion (SfM) techniques may produce sparse 3D point clouds when low-resolution input images are used, resulting in a low-quality mesh generation. Traditional point cloud upsampling techniques that improve the 3D point cloud resolution typically work on LiDAR-generated point clouds devoid of color information. Furthermore, most learned point cloud upsampling techniques compute graph features that capture local information by identifying a local neighborhood in a limited region around a point and hence may result in sub-optimal representation. To address these limitations, we propose CloudUP , a colored 3D point cloud upsampling approach that utilizes multi-scale spatial attention. Specifically, we design a novel across-scale attention mechanism for computing Multi-scale Point-Cloud Features (MPF) that capture 3D shape information of the point with respect to its neighborhood. We further extract spatial neighborhood-guided color features used to predict the color for the upsampled points. The color prediction is trained with a content-preserving loss function that aims to maintain intricate details and vivid colors. Our color refinement pipeline is guided by a vibrant colored dataset (collected by us) to assist in preserving the 3D contents.

CNN-based absolute camera pose estimation methods lack scene generalizability as the network is trained with scene-specific parameters. In this paper, we aim to solve the scene generalizability problem in 6-DoF camera pose estimation using a novel deep photo-geometric loss. We train a CNN-based relative pose estimation network end-to-end, by jointly optimizing the proposed deep photo-geometric loss along with the pose regression loss. Most traditional pose estimation methods use local keypoints to find 2D-2D correspondences, which fails under occlusion, textureless surfaces, motion blur, or repetitive structures. Given camera intrinsics, poses and depth, our method generates uniform 2D-2D photometric correspondence pairs via epipolar geometry during the training process with constraints to avoid textureless surfaces and occlusion, without the need of manually annotated keypoints information. The network is then trained with the correspondences information in such a way that not only the network learns from auxiliary photometric consistency information but also efficiently leverages scene geometry, consequently, we call it photo-geometric loss. The input to the photo-geometric loss layer is taken from the activation maps of the deep network, which contains much more information than a simple 2D-2D correspondence, and thus alleviating the need to choose a robust pose regression loss and its hyperparameters. With extensive experiments on three public datasets, we show that the proposed method significantly outperforms state-of-the-art relative pose estimation methods. The presented method also depicts state-of-the-art results on these datasets under cross-database evaluation settings, which proves its significance in terms of scene generalization.