Reconstructing Unsteady Flow Data From Representative Streamlines via Diffusion and Deep-Learning-Based Denoising

DL4SciVis: A State-of-the-Art Survey on Deep Learning for Scientific Visualization

“…We also refer to their research tasks when categorizing the surveyed papers in the respective tables according to learning type, network architecture, loss function, and evaluation metric. The description [51] TSR-TVD TVCG Han and Wang [50] SSR-TVD TVCG Han et al [55] STNet TVCG Wurster et al [167] arXiv Guo et al [47] SSR-VFD PVIS Jakob et al [76] TVCG Sahoo and Berger [126] IA-VFS EVIS An et al [2] STSRNet CG&A Han and Wang [53] TSR-VFD C&G Xie et al [168] tempoGAN TOG Werhahn et al [162] CGIT Wang et al [156] DeepOrganNet TVCG Lu et al [109] neurcomp CGF Weiss et al [160] fV-SRN arXiv Shi et al [131] GNN-Surrogate TVCG Han and Wang [54] VCNet VI Liu et al [106] JOV Han et al [49] CG&A Gu et al [45] VFR-UFD CG&A Han et al [56] V2V TVCG Gu et al [46] Scalar2Vec PVIS Kim et al [84] Deep Fluids CGF Chu et al [27] TOG Wiewel et al [163] LSP CGF Wiewel et al [164] LSS CGF Berger et al [12] TVCG Hong et al [70] DNN-VolVis PVIS He et al [63] InSituNet TVCG Weiss et al [159] TVCG Weiss et al [161] TVCG Weiss and Navab [158] DeepDVR arXiv He et al [62] CECAV-DNN VI Tkachev et al [143] TVCG Hong et al [71] PVIS Kim and Günther [85] CGF Han et al [57] arXiv Yang et al [169] JOV Shi and Tao [130] TIST Engel and Ropinski …”

“…This is not surprising as scalar field data are most commonly produced and widely available in SciVis. Apart from scalar field data, more than a quarter of the papers tackle vector field data, include 2D and 3D steady (e.g., [11], [49]) and unsteady (e.g., [45], [85]) vector fields. Two papers (i.e., [46], [52]) cover scalar and vector domains.…”

“…Data generation tasks produce or reconstruct data or models from low-resolution versions (e.g., [182]), feature representations (e.g., [49]), or lower-dimensional counterparts (e.g., [156]) [ ]. They also address data translation (e.g., [56]) [ ] (e.g., translating from one variable sequence to another or from an input field to another that satisfies specific properties or constraints) and extrapolation (e.g., [163]) [182] low-resolution volume super-resolution volume Han and Wang [51] TSR-TVD two end volumes intermediate volumes Han and Wang [50] SSR-TVD low-resolution volume super-resolution volume Han et al [55] STNet low-resolution two end volumes super-resolution intermediate volumes Wurster et al [167] low-resolution volume multi-resolution super-resolution volume Guo et al [47] SSR-VFD low-resolution field super-resolution field Jakob et al [76] low-resolution flow map super-resolution flow map Sahoo and Berger [126] IA-VFS low-resolution field super-resolution field An et al [2] STSRNet low-resolution two end fields super-resolution intermediate fields Han and Wang [53] TSR-VFD two end volumes intermediate volumes Xie et al [168] tempoGAN low-resolution volume super-resolution volume Werhahn et al [162] low-resolution volume super-resolution volume Wang et al [156] DeepOrganNet 3D/4D-CT projection or X-ray image 3D/4D lung model Lu et al [109] neurcomp voxel coordinate scalar value Weiss et al [160] fV-SRN voxel coordinate density or color Shi et al [131] GNN-Surrogate simulation parameters output field with adaptive resolution Han and Wang [54] VCNet incomplete volume complete volume Liu et al [106] volume patch feature vector Han et al [49] low-resolution field high-resolution field Gu et al [45] VFR-UFD low-quality fields high-quality fields Han et al [56] V2V source variable target variable Gu et al [46] Scalar2Vec scalar field volume vector field volume Kim et al [84] Deep Fluids velocity vector field, solver parameters divergence-free velocity field Chu et al…”

“…Their solution encompasses two steps: initializing a low-resolution field based on the input streamlines and upscaling the low-resolution field using a CNN. Gu et al [45] extended the work of Han et al [49] to reconstruct unsteady vector fields from their streamline representations via diffusion and DL-based denoising. Their solution captures temporal coherence by considering multiple consecutive timesteps at a time and preserves spatial coherence through streamline-based network optimization.…”

“…DL techniques can bring crucial benefits to SciVis. For example, inference can be performed more efficiently than conventional methods once a neural network is trained [45]. Furthermore, DL solutions offer a performance boost, such as data interpolation quality [55], reduction rate [109], or segmentation accuracy [154], compared with non-DL solutions.…”