Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HiFIVE: High-Fidelity Vector-Tile Reduction for Interactive Map Exploration

Tarlan Bahadori, Ahmed Eldawy

TL;DR

HiFIVE introduces a two-stage solution combining triage and sparsification to selectively prune records, attributes, and values based on information-theoretic and spatial criteria, and proves its NP-hardness.

Abstract

Interactive visualization is a common tool for exploring large open-data repositories, where users quickly explore datasets across diverse domains. When it comes to large-scale spatial data, many existing tools rely on server-side rendering to produce small images that can be viewed at the client-side. However, most users prefer client-side rendering that allows quick styling of the data for better visualization experience. This paper presents HiFIVE, a data-management framework for scalable, high-fidelity client-side geospatial visualization. We formalize the visualization-aware tile reduction problem, which captures the trade-off between tile-size and visualization distortion, and prove its NP-hardness. HiFIVE introduces a two-stage solution combining triage and sparsification to selectively prune records, attributes, and values based on information-theoretic and spatial criteria. Experiments demonstrate substantial tile-size reductions while preserving visual fidelity and interactive performance at terabyte scale.

HiFIVE: High-Fidelity Vector-Tile Reduction for Interactive Map Exploration

TL;DR

HiFIVE introduces a two-stage solution combining triage and sparsification to selectively prune records, attributes, and values based on information-theoretic and spatial criteria, and proves its NP-hardness.

Abstract

Interactive visualization is a common tool for exploring large open-data repositories, where users quickly explore datasets across diverse domains. When it comes to large-scale spatial data, many existing tools rely on server-side rendering to produce small images that can be viewed at the client-side. However, most users prefer client-side rendering that allows quick styling of the data for better visualization experience. This paper presents HiFIVE, a data-management framework for scalable, high-fidelity client-side geospatial visualization. We formalize the visualization-aware tile reduction problem, which captures the trade-off between tile-size and visualization distortion, and prove its NP-hardness. HiFIVE introduces a two-stage solution combining triage and sparsification to selectively prune records, attributes, and values based on information-theoretic and spatial criteria. Experiments demonstrate substantial tile-size reductions while preserving visual fidelity and interactive performance at terabyte scale.
Paper Structure (30 sections, 1 theorem, 33 equations, 10 figures, 9 tables, 1 algorithm)

This paper contains 30 sections, 1 theorem, 33 equations, 10 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Overview
  4. Visualization-Aware Tile Reduction Problem Definition
  5. Interpretation.
  6. MILP-based Sparsification
  7. Decision variables.
  8. Structural constraints.
  9. Linear size model.
  10. Objective function.
  11. Record utility.
  12. Cell utility.
  13. Visualization-Aware Triage
  14. Record-Level Triage
  15. Column-Level Triage
  16. ...and 15 more sections

Key Result

theorem 1

The visualization-aware tile reduction problem defined in eq:tile-reduction is NP-hard.

Figures (10)

  • Figure 1: Different client-side styles applied to the same set of vector tiles.
  • Figure 2: This is an overview of the HiFIVE pipeline.
  • Figure 3: Rasterization example of the tile $T_\text{in}$. (a) The input geometries. (b) The rasterization of the geometries. (c) The rasterization of salinity attribute ($j=3$). The empty cells denote $\bot$.
  • Figure 4: Examples of visualization stylings used in our experiments. We show two stylings for eBird (gradient based on observation time of the day and categorical based on country code), two for Roads (gradient based on length and categorical based on state), and two for Counties (gradient based on area of water and categorical based on state).
  • Figure 5: Scaling comparison across eBird subsets (8k to 801M records), roads subsets (7k to 70M records), and buildings subsets (45k to 455M records). Both axes are logarithmic.
  • ...and 5 more figures

Theorems & Definitions (12)

  • definition 1: Schema ($\boldsymbol{S}$)
  • definition 2: Feature
  • definition 3: Tile
  • definition 4: Column Size
  • definition 5: Tile Size
  • definition 6: Rendering Function ($\mathcal{R}$)
  • definition 7: Attribute Image
  • definition 8: Pixel-weighted Empirical Distribution
  • definition 9: Visualization-aware Attribute Divergence
  • definition 10: Tile Distortion
  • ...and 2 more