SciVisAgentBench-tasks / statistics /SciVisAgentBench_Taxnomoy.txt
KuangshiAi
add statistics
d1b209f
raw
history blame
8.6 kB
Data Taxonomy (several cross-cutting categories):
* Application Domains (this shouldn’t necessarily be a category rather a way to indicate the source of the data)
* Climate data
* SEM image data
* CT scan for objects
* Medical/CT/MRI scan
* Simulation data
* Molecular data
* ….
* Data Type
* Structured Data
* Image data (uniform rectilinear grids)
* Rectilinear grids
* Structured grids (curvilinear)
* AMR (Adaptive Mesh Refinement)?
* Unstructured Data
* Unstructured grids (arbitrary cell types)
* Polygonal data (surfaces, meshes)
* Point clouds
* Specialized Types
* Hyper-tree grids
* Composite/multi-block datasets
* Graph/network data
* Temporal Dimension
* Static/Single-Timestep
* Analysis of a single snapshot
* Spatial patterns only
* Time-Series
* Multiple discrete steps
* Temporal sequences
* Attribute Types (how the data is represented, what information it has)
* Scalar Fields (single value per point/cell)
* Temperature, pressure, density
* Scalars driving isosurfaces, color mapping
* Vector Fields (3-component direction/magnitude)
* Velocity, force, displacement
* Visualized via streamlines, glyphs, and LIC
* Tensor Fields (matrix at each point)
* Stress, strain, diffusion tensors
* Visualized via tensor glyphs, eigenvalue analysis
* Multi-variate/Multi-field/Multi-modal
* Multiple scalar/vector fields
* High-dimensional data visualization by analyzing multiple dimensions all at once
* Data Ensemble
* Multiple runs, parameter sweeps, etc.
Tasks Taxonomy:
Top level: 1) Atomic Operation, 2) Workflow, 3) Scientific Insights
1) Atomic Operation (individual well-defined operations):
* Extraction & Subsetting (Isolate regions of interest from larger datasets)
* Spatial/temporal Extraction
* Clipping (by plane, box, sphere, implicit function)
* Region selection (by point/cell IDs, bounding box)
* VOI extraction (volume of interest from structured data)
* Value-Based Selection
* Thresholding (scalar range filtering)
* Isocontouring/isosurfacing (constant value extraction)
* Connectivity filtering (connected region extraction)
* Sampling
* Subsampling/decimation (reduce resolution)
* Probing (sampling at specific points)
* Masking (regular/irregular point selection)
* Geometry & Topology Transformation (Change the structure or shape of data without necessarily changing attribute values)
* Geometric Modification
* Translation, rotation, scaling
* Deformation, warping
* Point/vertex manipulation
* Topological Changes
* Triangulation (convert polygons to triangles), Tessellation/subdivision
* Mesh refinement/coarsening
* Cell type conversion
* Boundary extraction (surface from volume)
* Structural Operations
* Merging datasets
* Appending data
* Splitting/partitioning
* Attribute Computation & Derivation (Calculate new data attributes from existing ones)
* Field Derivatives
* Gradient computation
* Divergence, curl, vorticity
* Curvature calculation
* Normal generation
* Scalar Operations
* Arithmetic operations on fields
* Vector magnitude computation
* Component extraction
* Field aggregation/statistics
* Advanced Computations[a]
* Tensor operations (eigenvalues, eigenvectors)
* Interpolation between fields
* Distance computations
* Time-Dependent Processing
* Temporal interpolation
* Particle tracing through time
* Flow integration [b](streamlines, pathlines, streaklines, timelines)
* Temporal statistics/aggregation
* Flow map computation
* Representation & Mapping (Transform data into visual representations)
* Glyph-Based Representation
* Oriented glyphs (arrows, cones)
* Scaled glyphs (size-mapped symbols)
* Tensor glyphs (ellipsoids, superquadrics)
* Volume markers
* Geometric Primitives
* Isosurfaces
* Contour lines/surfaces
* Cut planes/slices
* Ribbons, tubes, streamlines
* Color & Texture Mapping
* Scalar to color mapping
* Texture coordinate generation
* Opacity mapping
* Volume Representations
* Ray casting
* Splatting
* Smoothing & enhancement (Improve data quality or visual appearance)
* Smoothing Operations
* Surface smoothing (Laplacian, Gaussian)
* Data noise reduction
* Interpolation for filling gaps
* Enhancement
* Edge enhancement/detection
* Feature extraction
* Sharpening
* Filtering
* Outlier removal
* Statistical filtering
* View / Rendering Manipulation
* View Changes
* Rotation, zoom, move
* Render Options (should be belong to transformation)
* Light position
* Lighting mode
2) Workflow (sequence of operations with a clear visualization/analysis goal)
* Data Understanding & Exploration
* Data Characterization (statistics, distributions, quality check)
* Spatial Exploration (slicing, probing, overview generation)
* Feature Discovery (identifying interesting structures/patterns)
* Analysis & Quantification
* Statistical Analysis (descriptive statistics, distributions, correlations)
* Region-Based Measurement (volumes, areas, integrated quantities)
* Profile & Cross-Section Analysis (1D/2D extraction from 3D)
* Derived Quantity Computation (gradients, vorticity, custom fields)
* Feature Extraction & Tracking
* Structure Identification (vortices, shocks, boundaries, topology)
* Feature Characterization (properties, classification, quantification)
* Temporal Tracking (feature evolution, lifecycle, trajectories)
* Comparative & Temporal Analysis
* Multi-Variable Comparison (correlation, coordinated views)
* Temporal Evolution / Comparison (time-series comparison)
* Simulation Comparison (parameter studies, model validation)
* Difference Analysis (error fields, change detection)
* Flow & Transport Analysis[c]
* Trajectory Computation (streamlines, pathlines, streaklines)
* Lagrangian Analysis (particle tracking, residence time, FTLE)
* Transport Quantification (flux, mixing, coherent structures)
* Verification & Validation
* Data Quality Assessment (outliers, artifacts, boundary conditions)
* Code Verification (convergence, analytical comparison, consistency)
* Physical Validation (experimental comparison, uncertainty quantification)
* Data Processing & Optimization
* Data Conditioning (cleaning, smoothing, noise reduction)
* Data Reduction (decimation, sampling, compression)
* Format Conversion & Restructuring (mesh generation, type conversion)
* Parallel & Distributed Processing (HPC workflows, decomposition)[d]
* Uncertainty Quantification & Visualization[e]
* Sensitivity analysis, uncertainty characterization (estimate uncertainty)
* Aggregation and summarization (visualize quantile and interval)
* Communication & Dissemination
* Static Visualization (publication figures, high-resolution images)
* Animation Generation (temporal, spatial, parameter animations)
* Interactive Applications (web/desktop viewers, dashboards)
* Report Generation (automated analysis reports, summaries)
3) Scientific Insights (analysis or visualization that leads to domain-relevant insights)
* Application-specific questions and insights that can be derived from analysis or visualization
* The result could be: binary decisions, multiple choices,
* Potentially involves different workflow steps
What not to include
* Excessively large dataset (multiple GB to TB level)
* Interaction that ties to a specific tool/interface
* Questions do not have unique and clear answers/ground truth
[a]probably need better title
[b]Move flow integration into its own category of Vector Operations, analogous to Scalar Operations? Streamlines are not time-dependent. Also could include LIC, and maybe FTLE.
[c]There is overlap between this category and previous atomic operations on flow integration (see my earlier comment).
[d]Consider large scale HPC to be out of scope, in order to make the benchmark more broadly accessible?
[e]Group together with V&V?