Datasets:
Formats:
imagefolder
Size:
< 1K
ArXiv:
Tags:
visual-place-recognition
multimodal-learning
graph-neural-networks
urban-computing
pedestrian-navigation
day-night-recognition
License:
File size: 18,417 Bytes
e113db8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 | # Graph Structure Documentation
## Overview
The MMS-VPR dataset conceptualizes the pedestrian network as a spatial graph **G = (V, E)** where nodes represent street intersections and edges represent street segments. This graph-based organization provides:
1. **Structural clarity** for dataset navigation
2. **Foundation for GNN applications** (Graph Neural Networks)
3. **Spatial relationship modeling** for context-aware VPR
4. **Integration with urban science** through space syntax theory
The graph comprises **208 locations** organized into **four spatial types**:
- **81 Nodes**: Street intersections
- **125 Edges**: Pedestrian street segments (61 horizontal + 64 vertical)
- **2 Squares**: Large open spaces
## File Descriptions
### 00 Street Network Graph.pdf
**Visualization** of the complete pedestrian network showing:
- Geographic layout of all 208 locations
- Node positions (street intersections)
- Edge connections (street segments)
- Square boundaries
- Real-world mapping to Chengdu Taikoo Li
This map provides an intuitive understanding of:
- Spatial relationships between locations
- Network topology and connectivity
- Physical layout in the commercial district
**Use this visualization** to understand the overall dataset structure before diving into detailed data files.
---
### 01 Node Features.xlsx
**Contains**: Attributes for all 81 street intersection nodes (+ 2 squares)
#### Column Descriptions
| Column | Description | Example Values |
|--------|-------------|----------------|
| **Code** | Node identifier | `N-1-1`, `N-2-3`, `S-1` |
| **Node Type** | Classification (1-4) | `1`, `2`, `3`, `4` |
| **Location Coordinates** | Grid position (row, column) | `1, 1` → Row 1, Column 1 |
| **J-graph Depth** | Shortest path distance to public roads | `0`, `1`, `2`, `3`, `4`, `5` |
| **Longitude** | GPS longitude coordinate | `+104.081061` |
| **Latitude** | GPS latitude coordinate | `+30.658052` |
#### Node Type Classification
Nodes are classified based on **accessibility depth** (distance to public roads):
**Type 1: Gateway Nodes** (Depth 0-1)
- Located at or immediately adjacent to public roads
- Highest accessibility and visibility
- Serve as primary entry/exit points
- Characteristics: High pedestrian flow, easy wayfinding
**Type 2: Intermediate Nodes** (Depth 2-3)
- Moderately interior positions
- Balance between exposure and seclusion
- Act as conduits between entrances and deep zones
- Characteristics: Moderate flow, mixed accessibility
**Type 3: Deep Interior Nodes** (Depth 4-5)
- Deep interior locations
- Low visibility from public roads
- Lower pedestrian flow
- Higher wayfinding challenge
- Characteristics: Requires navigation through multiple turns
**Type 4: Open Squares**
- Large open spaces (S-1, S-2)
- Bounded by multiple edges and nodes
- Central gathering spaces
- Characteristics: High visibility within district, social hub
#### J-graph Depth Explanation
**Definition**: Shortest path distance (in number of street segments) from a node to the nearest public road.
- **Depth 0**: Node is ON a public road
- **Depth 1**: One street segment away from public road
- **Depth 2-3**: Interior positions requiring 2-3 turns from entrance
- **Depth 4-5**: Deep interior requiring complex navigation
This metric indicates:
- Ease of discovery by visitors
- Expected pedestrian traffic volume
- Wayfinding complexity
---
### 02 Edge Features.xlsx
**Contains**: Attributes for all 125 street segments (edges)
#### Column Descriptions
| Column | Description | Example Values |
|--------|-------------|----------------|
| **SegmentID** | Edge identifier | `Eh-1-1`, `Ev-2-3` |
| **NodeA** | Starting node | `N-1-1` |
| **NodeB** | Ending node | `N-1-2` |
| **Edge Type** | Street width classification (1-3) | `1`, `2`, `3` |
| **Location Coordinates** | Grid position | `1, 1` |
| **Orientation** | Street angle in degrees | `0` (horizontal), `90` (vertical) |
| **Length** | Physical length in meters | `53.05` |
| **Width** | Physical width in meters | `4.0` |
| **Urban Roads** | Public road indicator | `1` (public), `0` (private) |
| **Longitude** | GPS longitude of street center | `+104.082376` |
| **Latitude** | GPS latitude of street center | `+30.657317` |
| **e_loc_x** | First coordinate component | `1` |
| **e_loc_y** | Second coordinate component | `1` |
| **integration** | Space syntax integration (angular) | `0.015` |
| **betweenness** | Space syntax betweenness (angular) | `0.012` |
| **weighted_integration** | Weighted integration (angular + Euclidean) | `0.023` |
| **weighted_betweenness** | Weighted betweenness (angular + Euclidean) | `0.037` |
#### Edge Naming Convention
**Horizontal Edges** (East-West streets): `Eh-i-j`
- Example: `Eh-1-1` = Row 1, Column 1 of horizontal edges
- Orientation: 0°
**Vertical Edges** (North-South streets): `Ev-j-i`
- Example: `Ev-1-1` = Column 1, Row 1 of vertical edges
- Orientation: 90°
#### Edge Type Classification
Streets are classified by **width** following urban design principles:
**Type 1: Alley** (1-3 meters)
- Narrow passages
- Constrained flow corridors
- Primarily for connectivity
- Characteristics: Intimate scale, limited dwell potential
**Type 2: Intermediate Street** (4-7 meters)
- Comfortable circulation width
- Moderate dwell potential
- Supports walking + browsing
- Characteristics: Balanced flow and activity
**Type 3: Primary Broad Street** (8-13 meters)
- Wide open corridors
- Supports high pedestrian flow
- Suitable for events and gatherings
- Characteristics: High capacity, visual prominence
#### Urban Roads Indicator
- **1**: Public city road (district perimeter)
- **0**: Private interior street (within shopping district)
This distinguishes between:
- External connections to city street network
- Internal pedestrian-only circulation
#### Space Syntax Metrics
**Integration** (Global Accessibility)
- Measures how easily a street can be reached from all other streets
- **Higher values** → Centrally located, easily accessible streets
- **Lower values** → Peripheral or isolated streets
**Betweenness** (Through-Movement Potential)
- Measures how often a street lies on shortest paths between other streets
- **Higher values** → Primary routes with heavier pedestrian flow
- **Lower values** → Secondary streets with local traffic only
**Metric Variants**:
1. **integration / betweenness**: Based on **angular distance** only
- Angular distance = sum of turn angles / 90°
- Represents cognitive wayfinding effort
2. **weighted_integration / weighted_betweenness**: Based on **weighted distance**
- Weighted distance = 50% normalized Euclidean + 50% normalized angular
- Balances physical distance and turning complexity
- More comprehensive spatial measure
**Why Two Variants?**
- Angular metrics better predict pedestrian movement (people minimize turns)
- Weighted metrics balance physical distance with cognitive ease
- Different research questions may require different metrics
**Normalization**: All metrics are normalized to [0, 1] range for comparability.
---
### 03 Edge Connections.xlsx
**Contains**: Pairwise relationships between all connected street segments
#### Column Descriptions
| Column | Description | Example Values |
|--------|-------------|----------------|
| **Segment_X** | First street identifier | `Eh-1-1` |
| **Segment_Y** | Second street identifier | `Ev-1-1` |
| **SharedNode** | Intersection connecting X and Y | `N-1-1` |
| **Angle_X** | Orientation of street X | `0°` |
| **Angle_Y** | Orientation of street Y | `90°` |
| **TurnAngle** | Turn angle from X to Y | `90°` |
| **Angular Distance** | Normalized angular distance | `1.0` |
| **Euclidean Distance** | Physical distance (m) | `65.46` |
| **Weighted Distance** | Combined distance metric | `1.2` |
#### Distance Metrics Explained
**Turn Angle**
- Range: 0° to 180°
- Angle required to turn from street X to street Y
- Example: 90° = perpendicular turn, 180° = reverse direction
**Angular Distance**
- Formula: `TurnAngle / 90°`
- Range: 0.0 to 2.0
- Normalized cognitive cost of turning
- Example: 90° turn = 1.0, 45° turn = 0.5
**Euclidean Distance**
- Physical distance = Length(X) + Length(Y)
- Actual walking distance between street centers
- Unit: meters
**Weighted Distance**
- Combines angular and Euclidean distances
- Both components normalized to [0, 1] via min-max scaling
- Formula: `0.5 × norm(Euclidean) + 0.5 × norm(Angular)`
- Balances physical distance and turning difficulty
#### Use Cases
This file enables:
1. **Graph Construction**:
- Segment_X and Segment_Y define edge connectivity
- Build adjacency matrices for GNN input
2. **Pathfinding**:
- Use Angular Distance for cognitively easy routes
- Use Euclidean Distance for shortest physical paths
- Use Weighted Distance for balanced optimization
3. **Network Analysis**:
- Identify critical junctions (high-degree nodes)
- Analyze route options between locations
- Compute centrality measures
---
### 04 Square Features.xlsx
**Contains**: Attributes for the 2 large open square spaces
#### Column Descriptions
| Column | Description | Example Values |
|--------|-------------|----------------|
| **Square Code** | Square identifier | `S-1`, `S-2` |
| **Node Type** | Always 4 (open square) | `4` |
| **List of Adjacent Edges** | Boundary streets | `Eh-8-3, Ev-6-1, Eh-10-4` |
| **List of Adjacent Nodes** | Boundary intersections | `N-8-3, N-10-3, N-10-4` |
| **Size Rank** | Area ranking | `1` (largest), `2` |
| **J-graph Depth** | Minimum depth of adjacent nodes | `2` |
| **Longitude** | GPS longitude of square center | `+104.081061` |
| **Latitude** | GPS latitude of square center | `+30.658052` |
| **Perimeter** | Boundary length (m) | `198` |
| **Area** | Total area (m²) | `2347` |
#### Understanding Squares
Squares are **sub-graphs** defined as:
- **S_k = (E_k, V_k)** where:
- E_k ⊂ E: Subset of edges forming the boundary
- V_k ⊂ V: Subset of nodes at corners
**S-1: Central Square**
- Area: 2,347 m²
- Primary gathering space
- Highest visibility within district
- Multiple access points (edges)
**S-2: Secondary Square**
- Smaller auxiliary open space
- Complementary function to S-1
#### Adjacent Elements
**Adjacent Edges**: Streets forming the square's perimeter
- Define the boundary of the open space
- Connect the square to the rest of the network
**Adjacent Nodes**: Intersections at square corners
- Entry/exit points to the square
- Connection hubs to surrounding streets
#### J-graph Depth for Squares
- Calculated as the **minimum depth** among all adjacent nodes
- Represents the easiest path to reach the square from public roads
- Example: If adjacent nodes have depths [2, 3, 3, 4], square depth = 2
---
## Using Graph Structure in Research
### For GNN Applications
**Step 1: Build Adjacency Matrix**
```python
# Use 03 Edge Connections.xlsx
adjacency_matrix[Segment_X][Segment_Y] = 1 # Or use weighted distance
```
**Step 2: Extract Node Features**
```python
# Use 01 Node Features.xlsx and 02 Edge Features.xlsx
node_features = [depth, coordinates, integration, betweenness, ...]
```
**Step 3: Define Graph**
```python
G = (V, E, X)
V = all nodes (from 01)
E = all connections (from 03)
X = node/edge features (from 01, 02)
```
### For Spatial Analysis
**Accessibility Studies**:
- Use `integration` to identify most accessible streets
- Compare gateway nodes vs. deep interior nodes
- Analyze pedestrian flow distribution
**Route Planning**:
- Use `Angular Distance` for intuitive routes
- Use `Weighted Distance` for balanced optimization
- Consider `Edge Type` for comfort (prefer wider streets)
**Urban Design Insights**:
- Correlate `betweenness` with observed pedestrian counts
- Identify under-utilized spaces (low integration + low betweenness)
- Suggest interventions to improve connectivity
### For VPR Research
**Context-Aware Retrieval**:
- Use graph structure to constrain search space
- Prioritize neighbors in graph when matching queries
- Weight matches by spatial proximity in network
**Hierarchical Search**:
- First: Identify region using `Node Type` and `J-graph Depth`
- Then: Refine within local graph neighborhood
- Finally: Match specific location
**Flow-Aware Methods**:
- Weight locations by `integration` (high-traffic areas)
- Account for `betweenness` in query distribution
- Model expected pedestrian paths
---
## Space Syntax Theory Integration
### Background
**Space Syntax** is an established theory in urban science and architecture that quantifies spatial configuration's impact on human movement and social behavior.
**Key Insight**: The way spaces are connected (topology) influences how people move through them, independent of function or aesthetics.
### Integration Metric
**Mathematical Definition**:
$$\text{Integration}_i = \frac{k-2}{2 \left( \frac{1}{k-1} \sum_{j \neq i} d(i,j) - 1 \right)}$$
Where:
- k = total number of streets
- d(i,j) = shortest path distance from street i to street j
**Interpretation**:
- **High integration** → Street is central, easily reached from everywhere
- **Low integration** → Street is peripheral, requires many turns to access
**Practical Meaning**:
- Predicts pedestrian flow volume
- Indicates commercial potential
- Guides wayfinding difficulty
### Betweenness Metric
**Mathematical Definition**:
$$\text{Betweenness}_i = \sum_{\substack{j \neq i \\ k \neq i, k \neq j}} \frac{\sigma_{jk}(i)}{\sigma_{jk}}$$
Where:
- σ_jk = total number of shortest paths from j to k
- σ_jk(i) = number of those paths passing through i
**Interpretation**:
- **High betweenness** → Street is on many shortest paths (through-route)
- **Low betweenness** → Street is rarely on shortest paths (destination only)
**Practical Meaning**:
- Identifies primary circulation routes
- Predicts encounter probability
- Indicates strategic retail locations
### Why This Matters for VPR
Traditional VPR datasets provide only:
- Visual appearance
- GPS coordinates
MMS-VPR adds:
- **Topological context**: How is this place connected?
- **Movement potential**: How much traffic does it get?
- **Accessibility hierarchy**: How easy is it to find?
This enables:
1. **Context-aware matching**: Use spatial configuration to improve accuracy
2. **Query distribution modeling**: Account for non-uniform real-world usage
3. **Cross-city transfer**: Spatial patterns generalize across urban contexts
4. **Interpretable models**: Explain predictions using urban design theory
---
## Data Quality and Validation
### Topology Validation
✅ **Verified Properties**:
- All edges connect exactly two nodes
- No isolated components (graph is connected)
- Coordinates match real-world positions
- Angular distances computed correctly
### Metric Computation
✅ **Computation Tools**:
- Space syntax metrics: Computed using DepthmapX/UCL software
- Angular distance: Based on street centerline geometry
- Euclidean distance: Direct calculation from GPS + street length
✅ **Normalization**:
- All spatial metrics: Min-max normalized to [0, 1]
- Enables fair comparison across different distance types
---
## Frequently Asked Questions
**Q: Why use graph structure for VPR?**
A: Graph structure captures spatial relationships that pure visual matching misses. It enables context-aware retrieval, handles ambiguous queries better, and improves performance in visually similar areas.
**Q: What's the difference between angular and weighted distance?**
A: Angular distance considers only turns (cognitive ease). Weighted distance balances turns and physical distance (50-50). Choose based on your application: navigation apps may prefer angular, while distance estimation needs weighted.
**Q: How do I convert location codes to indices?**
A: Each location has a unique index 0-207:
- Nodes: N-1-1 to N-9-9 → indices 0-80
- Squares: S-1, S-2 → indices 81-82
- Horizontal edges: Eh-1-1 to Eh-11-7 → indices 83-143
- Vertical edges: Ev-1-1 to Ev-7-11 → indices 144-207
See `Texts/Annotations.xlsx` for complete mapping.
**Q: Can I use only visual data without graphs?**
A: Yes! The dataset works perfectly for standard image-based VPR. Graph structure is an **additional resource** for advanced methods, not a requirement.
---
## Recommended Workflow
### For First-Time Users
1. **Start** with `00 Street Network Graph.pdf` → Understand spatial layout
2. **Read** `Texts/Annotations.xlsx` → Learn location codes
3. **Explore** one location folder in `Images/` → See actual data
4. **Review** `01 Node Features.xlsx` → Understand location attributes
5. **Try** basic image retrieval → Validate setup
### For GNN Researchers
1. **Load** `03 Edge Connections.xlsx` → Build adjacency matrix
2. **Extract** features from `01` and `02` → Create feature vectors
3. **Align** with visual features from `Images/` → Multimodal embedding
4. **Train** GNN model → Leverage graph structure
5. **Evaluate** on graph-aware metrics → Consider topological accuracy
### For Urban Analytics
1. **Analyze** `integration` and `betweenness` → Identify key streets
2. **Correlate** with image counts → Validate with actual activity
3. **Compare** `J-graph Depth` across locations → Accessibility patterns
4. **Visualize** using `00 Street Network Graph.pdf` → Present findings
---
## Version History
**v1.0** (February 2026)
- Initial release
- 208 locations (81 nodes + 125 edges + 2 squares)
- Complete space syntax metrics
- Validated graph topology
---
## Citation
When using graph structure data, please cite:
```bibtex
@article{ou2025mmsvpr,
title = {MMS-VPR: Multimodal Street-Level Visual Place Recognition Dataset and Benchmark},
author = {Ou, Yiwei and Ren, Xiaobin and Sun, Ronggui and Gao, Guansong and
Jiang, Ziyi and Zhao, Kaiqi and Manfredini, Manfredo},
journal = {arXiv preprint arXiv:2505.12254},
year = {2025},
url = {https://arxiv.org/abs/2505.12254}
}
```
## References
For more on space syntax theory:
- Hillier, B., & Hanson, J. (1984). The Social Logic of Space
- Turner, A. (2001). Angular analysis. 3rd International Symposium on Space Syntax
- Peponis, J., et al. (1997). The spatial core of urban culture
---
**For questions about graph structure, please open an issue on GitHub or contact the authors.**
|