Duplicate from wanchun-ni/STRIDE-Bench

Co-authored-by: sweetspot00 <wanchun-ni@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,60 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.avro filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mds filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text
+# Video files - compressed
+*.mp4 filter=lfs diff=lfs merge=lfs -text
+*.webm filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,246 @@
+---
+license: mit
+task_categories:
+  - text-generation
+  - question-answering
+language:
+  - en
+tags:
+  - crowd-simulation
+  - trajectory-evaluation
+  - pedestrian-dynamics
+  - benchmark
+  - behavioral-evaluation
+pretty_name: STRIDE-Bench
+size_categories:
+  - 1K<n<10K
+configs:
+  - config_name: benchmark
+    data_files: benchmark.jsonl
+  - config_name: scenes
+    data_files: scenes.parquet
+  - config_name: maps
+    data_files: maps.parquet
+---
+
+# STRIDE Benchmark
+
+## Dataset Description
+
+STRIDE Benchmark is an evaluation benchmark for assessing the behavioral realism of crowd trajectory generation and simulation models. Rather than comparing trajectories point-by-point, it evaluates whether generated trajectories exhibit behaviors consistent with a given scenario description — measuring **trajectory-context consistency** through decomposed behavioral questions.
+
+### Dataset Summary
+
+The dataset is distributed as three files, all aligned by scene:
+
+| File | Format | Rows | Purpose |
+|------|--------|------|---------|
+| `benchmark.jsonl` | JSON Lines | 936 | Behavioral evaluation specs (questions + measurements per scene) |
+| `scenes.parquet`  | Parquet    | 936 | Scenario metadata, initial agent states, group structure, goals |
+| `maps.parquet`    | Parquet    | 936 | Per-scene obstacle rectangles + rendered obstacle PNG |
+
+| Property | Value |
+|----------|-------|
+| Scenes | 936 |
+| Total Questions | 6,633 |
+| Total Measurements | 11,696 |
+| Evaluation Functions | 22 |
+| Behavioral Categories | 11 |
+| Unique Scenarios | 31 |
+| Locations | 30 real-world locations |
+
+### Languages
+
+English
+
+## Dataset Structure
+
+The dataset has three configs — `benchmark`, `scenes`, `maps` — each backed by one file. All three are aligned: row `i` in any file corresponds to the same scene, joinable on `scene_id` / `source_scene_id`.
+
+### `benchmark.jsonl` — behavioral evaluation specs
+
+Each line is one scene. Fields:
+
+- **`source_scene_id`** (`string`): Unique scene identifier.
+- **`scenario_id`** (`string`): Reference to the physical scenario / environment.
+- **`description`** (`string`): Natural-language description of the crowd behavior scenario.
+- **`category`** (`string`): Behavioral category label (one of 11).
+- **`location`** (`string`): Real-world location name.
+- **`decomposition_reasoning`** (`string`): LLM reasoning for how the description maps to measurable behavioral properties.
+- **`questions`** (`list`): Behavioral evaluation questions, each containing:
+  - `id` (`string`): Question identifier (e.g., "Q1").
+  - `question` (`string`): Natural-language behavioral question.
+  - `measurements` (`list`): Quantitative measurements to answer the question:
+    - `function` (`string`): Name of the evaluation function.
+    - `params` (`object`): Function parameters.
+    - `expected_result` (`object`): Expected value range with `min` and/or `max` bounds.
+    - `expected_result_reasoning` (`string`): Justification for the expected range.
+    - `delta_value` (`number`): Tolerance/margin for soft evaluation.
+- **`_model`** (`string`): The LLM used for benchmark generation (carried per-row from the original top-level `model` field).
+- **`_n_scenes`** (`int`): Total number of scenes in the dataset (carried per-row).
+
+### `scenes.parquet` — scene state and metadata
+
+One row per scene. Selected columns (full schema in the Parquet file):
+
+- **`scene_id`** (`string`): Scene identifier (joins with `benchmark.source_scene_id` and `maps.scene_id`).
+- **`scene_index`** (`int`): Numeric index of the scene.
+- **`scenario`** (`string`): Free-form scenario description.
+- **`category`** (`string`): Behavioral category label.
+- **`crowd_size`** (`int`), **`crowd_size_label`** (`string`): Number of agents and bucketed label.
+- **`ungrouped_agents`** (`int`): Count of agents not in a group.
+- **`event_center_px`** / **`event_center_m`** (`list<float>`): Event center in pixel and meter coordinates.
+- **`initial_state`** (`list<list<float>>`): Per-agent starting state.
+- **`groups`** (`list<list<int>>`): Agent grouping (each inner list is one group's member indices).
+- **`towards_event`**, **`towards_goal`** (`bool`): Whether agents move toward the event / goal.
+- **`desired_speed_range`**, **`collision_rate_early`**, **`collision_rate_late`**, **`relavance_to_event_range`**, **`lingering_fraction_range`** (`list<float>`): Expected behavioral ranges used during simulation.
+- **`goal_location_raw`**, **`goals_px`**, **`goals_m`**, **`goal_location_px`**, **`goal_location_m`** (`string`): Goal specifications. Stored as JSON-encoded strings because their runtime types vary across scenes (string / list / dict / null).
+- **`assets`** (`struct`): Paths/refs to anchored obstacles, homography, and obstacle PNG.
+
+### `maps.parquet` — per-scene obstacle data
+
+One row per scene. Columns:
+
+- **`scene_id`** (`string`): Scene identifier (joins with the other tables).
+- **`obstacles_key`** (`string`): Original npz key (always `"obstacles"`).
+- **`obstacles`** (`list<list<float>>`): Obstacle rectangles, shape `(N, 4)` — the contents of `map.npz["obstacles"]` from the source layout.
+- **`obstacles_shape`** (`list<int>`): Original array shape, for reconstruction.
+- **`obstacle_map_png`** (`bytes`): Raw PNG bytes of the rendered obstacle map.
+
+### Behavioral Categories
+
+| Category | Count | Description |
+|----------|-------|-------------|
+| Escaping | 123 | Emergency/evacuation behavior |
+| Violent | 122 | Conflict scenarios |
+| Demonstrator | 106 | Organized group/protest movement |
+| Dense | 103 | High-density crowd movement |
+| Aggressive | 89 | Aggressive crowd behavior |
+| Rushing | 85 | Fast-paced movement |
+| Expressive | 76 | Emotional/expressive movement |
+| Participatory | 75 | Event-based/interactive |
+| Cohesive | 67 | Group-walking behavior |
+| Ambulatory | 58 | Normal walking/commuting |
+| Disability | 32 | Mobility-constrained movement |
+
+### Evaluation Functions (20 total)
+
+
+**V-Velocy:**
+- `mean_speed` — Average walking speed (m/s)
+- `speed_variation_coeff` — Speed heterogeneity (coefficient of variation)
+
+**R-Realism:**
+- `collision_fraction` — Fraction of near-collision events
+- `lingering_fraction` — Fraction of slow/stationary agents
+
+
+**D-Direction:**
+- `flow_alignment` — Movement directionality alignment (0–1)
+- `path_linearity` — Straightness of paths (0–1)
+- `directional_entropy_normalized` — Direction diversity (0–1)
+
+**S-Spatial:**
+- `spatial_concentration` — Non-uniform spatial distribution (Gini-based, 0–1)
+- `mean_local_density` — Average neighbor count within adaptive radius
+- `peak_local_density` — Maximum local density (hotspot intensity)
+- `dispersal_score` — Radial divergence from initial centroid
+- `convergence_score` — Inward convergence toward final centroid
+
+
+**T-Temporal:**
+- `clustering_trend` — Group formation dynamics
+- `lingering_trend` — Agents stopping/mobilizing over time
+- `entropy_trend` — Movement organization vs. chaos over time
+- `speed_trend` — Temporal acceleration/deceleration pattern
+- `density_trend` — Crowd compacting or dispersing over time
+- `collision_trend` — Collision escalation/resolution over time
+- `flow_alignment_trend` — Movement becoming organized or chaotic
+
+### Example
+
+One row of `benchmark.jsonl`:
+
+```json
+{
+  "source_scene_id": "0000",
+  "scenario_id": "00_Zurich_HB_simplified_obstacle",
+  "description": "Morning commuter flow at Zurich HB with people casually walking between ticket machines and platforms, navigating the concourse and open corridors.",
+  "category": "Ambulatory",
+  "location": "Zurich",
+  "questions": [
+    {
+      "id": "Q1",
+      "question": "Are pedestrians moving at normal-but-constrained commuter walking speeds (not rushing/running)?",
+      "measurements": [
+        {
+          "function": "mean_speed",
+          "params": {},
+          "expected_result": {"min": 0.8, "max": 1.3},
+          "expected_result_reasoning": "Morning commuters are walking casually; in a station concourse with obstacles, average walking speed is typically 0.8–1.2 m/s.",
+          "delta_value": 0.07
+        }
+      ]
+    }
+  ],
+  "_model": "...",
+  "_n_scenes": 936
+}
+```
+
+## Usage
+
+Each file is exposed as a separate config. Load whichever you need:
+
+```python
+from datasets import load_dataset
+
+REPO = "anonymous1ads34/STRIDE-Bench"
+
+benchmark = load_dataset(REPO, "benchmark", split="train")
+scenes    = load_dataset(REPO, "scenes",    split="train")
+maps      = load_dataset(REPO, "maps",      split="train")
+
+print(benchmark[0]["description"])
+print(scenes[0]["scene_id"], scenes[0]["crowd_size"])
+print(maps[0]["scene_id"], maps[0]["obstacles_shape"])
+```
+
+Rows are aligned across the three configs and joinable on `scene_id` (or `source_scene_id` in `benchmark`).
+
+### Reconstructing maps from Parquet
+
+```python
+import io, numpy as np
+from PIL import Image
+
+row = maps[0]
+obstacles = np.array(row["obstacles"], dtype=np.float64).reshape(row["obstacles_shape"])
+img = Image.open(io.BytesIO(row["obstacle_map_png"]))
+```
+
+### Intended Use
+
+This benchmark is designed to evaluate crowd trajectory generation models by measuring whether their outputs exhibit behaviorally consistent movement patterns given a textual scenario description. It enables:
+
+- **Model comparison**: Rank trajectory generators by behavioral realism across diverse scenarios.
+- **Category-level analysis**: Identify model strengths/weaknesses per behavior type (e.g., evacuation vs. normal walking).
+- **Trajectory Question Answering (STRIDE)**: Frame trajectory evaluation as a QA task — does the trajectory "answer" behavioral questions correctly?
+
+### Evaluation Protocol
+
+For each model and scene:
+1. Generate/simulate a trajectory given the scene description.
+2. Execute evaluation functions on the generated trajectory.
+3. Compare actual function outputs to expected ranges (with `delta_value` tolerance).
+4. Compute pass/fail per question and aggregate STRIDE accuracy.
+
+### Trajectory Format
+
+Models are expected to produce trajectories as arrays of shape `(timesteps, agents, 7)` with columns: `[x, y, vx, vy, goal_x, goal_y, radius]`.
+
+## Limitations
+
+- Expected value ranges are LLM-estimated and may not perfectly match all real-world instances of a described scenario.
+- The benchmark focuses on aggregate behavioral statistics, not individual trajectory plausibility.
+- Location geometry (obstacles, boundaries) is simplified from real environments.

croissant/benchmark.croissant.json

@@ -0,0 +1,224 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "repeated": "cr:repeated",
+    "sc": "https://schema.org/",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform"
+  },
+  "@type": "sc:Dataset",
+  "name": "STRIDE-Bench-benchmark",
+  "description": "Behavioral evaluation specs for the STRIDE-Bench crowd trajectory benchmark. Each row is one scene with a natural-language scenario description and a decomposition into behavioral evaluation questions (Trajectory Question Answering / TQA), each grounded in one or more measurement functions with expected value ranges and tolerances.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "license": "https://choosealicense.com/licenses/mit/",
+  "url": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark",
+  "version": "1.0.0",
+  "datePublished": "2026-04-20",
+  "keywords": [
+    "crowd-simulation",
+    "trajectory-evaluation",
+    "pedestrian-dynamics",
+    "benchmark",
+    "behavioral-evaluation"
+  ],
+  "creator": {
+    "@type": "Person",
+    "name": "Anonymous (NeurIPS 2026 D&B submission, double-blind)"
+  },
+  "rai:dataCollectionType": [
+    "Synthetic",
+    "Simulation"
+  ],
+  "rai:personalSensitiveInformation": "None. The dataset contains no personal data, no images of identifiable individuals, and no real pedestrian tracks. All scenes are synthetic.",
+  "rai:hasSyntheticData": true,
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "benchmark-jsonl",
+      "name": "benchmark.jsonl",
+      "description": "JSON Lines: one scene per line. Each line contains the scene's scenario description, behavioral category, decomposition reasoning, and a list of behavioral evaluation questions with measurement specs.",
+      "contentUrl": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark/resolve/main/benchmark.jsonl",
+      "encodingFormat": "application/jsonlines",
+      "sha256": "8b44947d2e0281edfd2274ec403b7087af18e8ec1a04ffeb22837f1c90d52800",
+      "contentSize": "7228940 B"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "benchmark",
+      "name": "benchmark",
+      "description": "One record per scene (one JSONL line).",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/source_scene_id",
+          "name": "source_scene_id",
+          "description": "Unique scene identifier.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.source_scene_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/scenario_id",
+          "name": "scenario_id",
+          "description": "Reference to the physical scenario / environment.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.scenario_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/description",
+          "name": "description",
+          "description": "Natural-language description of the crowd behavior scenario.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.description"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/category",
+          "name": "category",
+          "description": "Behavioral category label (one of 11).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.category"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/location",
+          "name": "location",
+          "description": "Real-world location name.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.location"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/decomposition_reasoning",
+          "name": "decomposition_reasoning",
+          "description": "LLM reasoning for description -> behavioral properties.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.decomposition_reasoning"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/questions",
+          "name": "questions",
+          "description": "Nested list of {id, question, measurements:[{function, params, expected_result, expected_result_reasoning, delta_value}]}.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.questions"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/_model",
+          "name": "_model",
+          "description": "LLM used to generate the benchmark (carried per-row).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$._model"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/_n_scenes",
+          "name": "_n_scenes",
+          "description": "Total scenes in the dataset (carried per-row).",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$._n_scenes"
+            }
+          }
+        }
+      ]
+    }
+  ]
+}

croissant/maps.croissant.json

@@ -0,0 +1,164 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "repeated": "cr:repeated",
+    "sc": "https://schema.org/",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform"
+  },
+  "@type": "sc:Dataset",
+  "name": "STRIDE-Bench-maps",
+  "description": "Per-scene map data for STRIDE-Bench: obstacle rectangles (originally stored as map.npz with key 'obstacles', shape (N, 4)) and a rendered obstacle PNG, packed into a single Parquet table with one row per scene.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "license": "https://choosealicense.com/licenses/mit/",
+  "url": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark",
+  "version": "1.0.0",
+  "datePublished": "2026-04-20",
+  "keywords": [
+    "crowd-simulation",
+    "trajectory-evaluation",
+    "pedestrian-dynamics",
+    "benchmark",
+    "behavioral-evaluation"
+  ],
+  "creator": {
+    "@type": "Person",
+    "name": "Anonymous (NeurIPS 2026 D&B submission, double-blind)"
+  },
+  "rai:dataCollectionType": [
+    "Synthetic",
+    "Simulation"
+  ],
+  "rai:personalSensitiveInformation": "None. The dataset contains no personal data, no images of identifiable individuals, and no real pedestrian tracks. All scenes are synthetic.",
+  "rai:hasSyntheticData": true,
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "maps-parquet",
+      "name": "maps.parquet",
+      "description": "Parquet table of per-scene map data: obstacle rectangles (N x 4 float64) and a rendered obstacle PNG. One row per scene.",
+      "contentUrl": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark/resolve/main/maps.parquet",
+      "encodingFormat": "application/x-parquet",
+      "sha256": "1a15c3423dba429e9c805c05eb4afb42525efacce7931dd1991a05664cc0c6c4",
+      "contentSize": "7735446 B"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "maps",
+      "name": "maps",
+      "description": "One record per scene; obstacles array + rendered obstacle PNG.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "maps/scene_id",
+          "name": "scene_id",
+          "description": "scene_id (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "scene_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "maps/obstacles_key",
+          "name": "obstacles_key",
+          "description": "obstacles_key (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "obstacles_key"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "maps/obstacles",
+          "name": "obstacles",
+          "description": "obstacles (parquet type: list<element: list<element: double>>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "obstacles"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "maps/obstacles_shape",
+          "name": "obstacles_shape",
+          "description": "obstacles_shape (parquet type: list<element: int64>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "obstacles_shape"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "maps/obstacle_map_png",
+          "name": "obstacle_map_png",
+          "description": "obstacle_map_png (parquet type: binary)",
+          "dataType": "sc:ImageObject",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "obstacle_map_png"
+            }
+          }
+        }
+      ]
+    }
+  ]
+}

croissant/rai/croissant_rai_STRIDE-Bench-benchmark.json

@@ -0,0 +1,239 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "repeated": "cr:repeated",
+    "sc": "https://schema.org/",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform",
+    "prov": "http://www.w3.org/ns/prov#"
+  },
+  "@type": "sc:Dataset",
+  "name": "STRIDE-Bench-benchmark",
+  "description": "Behavioral evaluation specs for the STRIDE-Bench crowd trajectory benchmark. Each row is one scene with a natural-language scenario description and a decomposition into behavioral evaluation questions (Trajectory Question Answering / TQA), each grounded in one or more measurement functions with expected value ranges and tolerances.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "license": "https://choosealicense.com/licenses/mit/",
+  "url": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark",
+  "version": "1.0.0",
+  "datePublished": "2026-04-20",
+  "keywords": [
+    "crowd-simulation",
+    "trajectory-evaluation",
+    "pedestrian-dynamics",
+    "benchmark",
+    "behavioral-evaluation"
+  ],
+  "creator": {
+    "@type": "Person",
+    "name": "Anonymous (NeurIPS 2026 D&B submission, double-blind)"
+  },
+  "rai:dataCollectionType": [
+    "Synthetic",
+    "Simulation"
+  ],
+  "rai:personalSensitiveInformation": "None. The dataset contains no personal data, no images of identifiable individuals, and no real pedestrian tracks. All scenes are synthetic.",
+  "rai:hasSyntheticData": true,
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "benchmark-jsonl",
+      "name": "benchmark.jsonl",
+      "description": "JSON Lines: one scene per line. Each line contains the scene's scenario description, behavioral category, decomposition reasoning, and a list of behavioral evaluation questions with measurement specs.",
+      "contentUrl": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark/resolve/main/benchmark.jsonl",
+      "encodingFormat": "application/jsonlines",
+      "sha256": "8b44947d2e0281edfd2274ec403b7087af18e8ec1a04ffeb22837f1c90d52800",
+      "contentSize": "7228940 B"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "benchmark",
+      "name": "benchmark",
+      "description": "One record per scene (one JSONL line).",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/source_scene_id",
+          "name": "source_scene_id",
+          "description": "Unique scene identifier.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.source_scene_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/scenario_id",
+          "name": "scenario_id",
+          "description": "Reference to the physical scenario / environment.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.scenario_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/description",
+          "name": "description",
+          "description": "Natural-language description of the crowd behavior scenario.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.description"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/category",
+          "name": "category",
+          "description": "Behavioral category label (one of 11).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.category"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/location",
+          "name": "location",
+          "description": "Real-world location name.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.location"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/decomposition_reasoning",
+          "name": "decomposition_reasoning",
+          "description": "LLM reasoning for description -> behavioral properties.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.decomposition_reasoning"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/questions",
+          "name": "questions",
+          "description": "Nested list of {id, question, measurements:[{function, params, expected_result, expected_result_reasoning, delta_value}]}.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$.questions"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/_model",
+          "name": "_model",
+          "description": "LLM used to generate the benchmark (carried per-row).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$._model"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "benchmark/_n_scenes",
+          "name": "_n_scenes",
+          "description": "Total scenes in the dataset (carried per-row).",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "benchmark-jsonl"
+            },
+            "extract": {
+              "jsonPath": "$._n_scenes"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "rai:dataLimitations": "This benchmark does not contain any real-world events.",
+  "rai:dataBiases": "There's a selection bias that we sample rarer scenarios such as Violent and Escaping, as they are underrepresented in the real world.",
+  "rai:dataUseCases": "This data is used for evaluating the alignment of text-to-trajectory generation. The input is the generated trajectory coordinates timeseries dataset.",
+  "rai:dataSocialImpact": "As this is to evaluate pedestrian trajectory generation, it would help in urban planning, venue safety auditing, and emergency preparedness. There's no obvious risk in using it.",
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Benchmark Generation",
+      "sc:description": "We curate 113 location maps of high-traffic public venues, including major tourist destinations, stadiums, university campuses, and transit hubs. These venues routinely host dense crowds during events. The maps are acquired through the Google Maps API and converted into obstacle representations through a three-stage pipeline: semantic segmentation based on the official Google Maps colormap, polygon approximation of segmented obstacle regions, and rasterization to a fixed $301.7~\\text{m} \\times 282.8~\\text{m}$ canvas. Figure~\\ref{fig:real-world-obstacle} illustrates this map-processing pipeline.\n\\textbf{Scenario descriptions.}\nWe adopt Berlonghi's typology of 11 crowd categories~\\cite{berlonghi1995understanding}. For each category, we prompt GPT-5.1 with a structured template that elicits: (i) a natural-language scenario description, (ii) candidate exits and event-center coordinates in the map pixel grid, and (iii) the initial crowd size and spatial distribution. Fields (ii) and (iii) provide a common initialization across models, allowing us to isolate context-conditioned behavior from setup variability. Prompts are provided in Appendix~\\ref{app:scenario-corpus}.\n\\textbf{Filtering and stratified sampling.}\nWe remove scenarios with insufficient agent density, exits or event centers that fall on obstacles, and near-duplicate descriptions. Obstacle conflicts are a known failure mode of polygon approximation, while duplicates are identified by cosine-similarity filtering over sentence embeddings. We also conduct a manual pass to remove residual map-realism failures.\nFrom the filtered pool, we perform \\emph{stratified oversampling} toward safety-critical categories, such as Escaping, Violent, and Dense. These categories are systematically underrepresented in real-world trajectory datasets~\\cite{pellegrini2009you, lerner2007crowds, robicquet2016learning}, since such dynamics cannot be ethically staged. They are also the regimes where context-conditioned generation has the highest applied value, including urban planning, venue safety auditing, and emergency preparedness. For each retained scenario, we prompt GPT-5.2 to generate the behavioral questions, measurements, and expected answers. Prompts are provided in Appendix~\\ref{app:qa-generation}.\n\\textbf{Calibration.}\nTo improve the human alignment of expected answers, we incorporate real-world crowd reference data into our human-trajectory knowledge base, \\emph{TrajFacts}. These references include events such as the Love Parade~\\cite{zhao2020assessing, krausz2012loveparade} and the Itaewon crowd crush~\\cite{hwang2025virtual}. \\emph{TrajFacts} provides reference values that help the model produce more realistic expected-answer ranges. Appendix~\\ref{app:gt-library} gives additional details.\nIn total, STRIDE-Bench contains 936 scenarios, 6,633 questions, and 11,696 measurements. Figure~\\ref{fig:benchmark-stats} shows the heatmap of DMT function usage across crowd categories.\n\\textbf{Usage.}\nSTRIDE supports two evaluation modes. In \\emph{benchmark mode}, evaluation uses the released scenarios together with cached questions and expected-answer ranges. The final score is therefore produced entirely by deterministic function calls, with no LLM required at scoring time. In \\emph{open mode}, an LLM agent generates the question set on the fly from a user-supplied scenario description, enabling evaluation beyond the released benchmark."
+    }
+  ]
+}

croissant/rai/croissant_rai_STRIDE-Bench-maps.json

@@ -0,0 +1,185 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "repeated": "cr:repeated",
+    "sc": "https://schema.org/",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform",
+    "prov": "http://www.w3.org/ns/prov#"
+  },
+  "@type": "sc:Dataset",
+  "name": "STRIDE-Bench-maps",
+  "description": "Per-scene map data for STRIDE-Bench: obstacle rectangles (originally stored as map.npz with key 'obstacles', shape (N, 4)) and a rendered obstacle PNG, packed into a single Parquet table with one row per scene.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "license": "https://choosealicense.com/licenses/mit/",
+  "url": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark",
+  "version": "1.0.0",
+  "datePublished": "2026-04-20",
+  "keywords": [
+    "crowd-simulation",
+    "trajectory-evaluation",
+    "pedestrian-dynamics",
+    "benchmark",
+    "behavioral-evaluation"
+  ],
+  "creator": {
+    "@type": "Person",
+    "name": "Anonymous (NeurIPS 2026 D&B submission, double-blind)"
+  },
+  "rai:dataCollectionType": [
+    "Synthetic",
+    "Simulation"
+  ],
+  "rai:personalSensitiveInformation": "None. The dataset contains no personal data, no images of identifiable individuals, and no real pedestrian tracks. All scenes are synthetic.",
+  "rai:hasSyntheticData": false,
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "maps-parquet",
+      "name": "maps.parquet",
+      "description": "Parquet table of per-scene map data: obstacle rectangles (N x 4 float64) and a rendered obstacle PNG. One row per scene.",
+      "contentUrl": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark/resolve/main/maps.parquet",
+      "encodingFormat": "application/x-parquet",
+      "sha256": "1a15c3423dba429e9c805c05eb4afb42525efacce7931dd1991a05664cc0c6c4",
+      "contentSize": "7735446 B"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "maps",
+      "name": "maps",
+      "description": "One record per scene; obstacles array + rendered obstacle PNG.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "maps/scene_id",
+          "name": "scene_id",
+          "description": "scene_id (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "scene_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "maps/obstacles_key",
+          "name": "obstacles_key",
+          "description": "obstacles_key (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "obstacles_key"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "maps/obstacles",
+          "name": "obstacles",
+          "description": "obstacles (parquet type: list<element: list<element: double>>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "obstacles"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "maps/obstacles_shape",
+          "name": "obstacles_shape",
+          "description": "obstacles_shape (parquet type: list<element: int64>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "obstacles_shape"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "maps/obstacle_map_png",
+          "name": "obstacle_map_png",
+          "description": "obstacle_map_png (parquet type: binary)",
+          "dataType": "sc:ImageObject",
+          "source": {
+            "fileObject": {
+              "@id": "maps-parquet"
+            },
+            "extract": {
+              "column": "obstacle_map_png"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "rai:dataLimitations": "The map is originally from real-world Google map, but simplified into simple polygon obstacles. It may not be used as a real-world map.",
+  "rai:dataBiases": "We sampled only public areas with a large crowd of people.",
+  "rai:dataUseCases": "This map is used as an input for text-to-trajectory generation. Obstacles could be given to the model for its understanding of the geometry context.",
+  "rai:dataSocialImpact": "As STRIDE-Bench is used for text-to-trajectory generation, it could facilitate in urban planning, venue safety auditing, and emergency preparedness.",
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://www.google.com/maps",
+      "prov:label": "Google Map"
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Maps",
+      "sc:description": "We curate 113 location maps of high-traffic public venues, including major tourist destinations, stadiums, university campuses, and transit hubs. These venues routinely host dense crowds during events. The maps are acquired through the Google Maps API and converted into obstacle representations through a three-stage pipeline: semantic segmentation based on the official Google Maps colormap, polygon approximation of segmented obstacle regions, and rasterization to a fixed $301.7~\\text{m} \\times 282.8~\\text{m}$ canvas. Figure~\\ref{fig:real-world-obstacle} illustrates this map-processing pipeline."
+    }
+  ]
+}

croissant/rai/croissant_rai_STRIDE-Bench-scenes.json

@@ -0,0 +1,464 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "repeated": "cr:repeated",
+    "sc": "https://schema.org/",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform",
+    "prov": "http://www.w3.org/ns/prov#"
+  },
+  "@type": "sc:Dataset",
+  "name": "STRIDE-Bench-scenes",
+  "description": "Crowd-simulation scenes for STRIDE-Bench: scenario metadata, initial agent states (position/velocity/goal/radius), group structure, and per-scene expected behavioral ranges. Distributed as a single Parquet table with one row per scene; nested fields whose runtime types vary (e.g. goal_location_*) are stored as JSON-encoded strings.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "license": "https://choosealicense.com/licenses/mit/",
+  "url": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark",
+  "version": "1.0.0",
+  "datePublished": "2026-04-20",
+  "keywords": [
+    "crowd-simulation",
+    "trajectory-evaluation",
+    "pedestrian-dynamics",
+    "benchmark",
+    "behavioral-evaluation"
+  ],
+  "creator": {
+    "@type": "Person",
+    "name": "Anonymous (NeurIPS 2026 D&B submission, double-blind)"
+  },
+  "rai:dataCollectionType": [
+    "Synthetic",
+    "Simulation"
+  ],
+  "rai:personalSensitiveInformation": "None. The dataset contains no personal data, no images of identifiable individuals, and no real pedestrian tracks. All scenes are synthetic.",
+  "rai:hasSyntheticData": true,
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "scenes-parquet",
+      "name": "scenes.parquet",
+      "description": "Parquet table of crowd-simulation scenes: scenario metadata, initial agent states, group structure, goal locations, and expected behavioral ranges. One row per scene.",
+      "contentUrl": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark/resolve/main/scenes.parquet",
+      "encodingFormat": "application/x-parquet",
+      "sha256": "b73edd5a977fc7525330a83ba8a61218953933febdd7de66b27116ff68bd835f",
+      "contentSize": "5641744 B"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "scenes",
+      "name": "scenes",
+      "description": "One record per scene; columns derived from the Parquet schema.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/scene_id",
+          "name": "scene_id",
+          "description": "scene_id (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "scene_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/scene_index",
+          "name": "scene_index",
+          "description": "scene_index (parquet type: int64)",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "scene_index"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/scenario",
+          "name": "scenario",
+          "description": "scenario (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "scenario"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/category",
+          "name": "category",
+          "description": "category (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "category"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/crowd_size_label",
+          "name": "crowd_size_label",
+          "description": "crowd_size_label (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "crowd_size_label"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/crowd_size",
+          "name": "crowd_size",
+          "description": "crowd_size (parquet type: int64)",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "crowd_size"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/ungrouped_agents",
+          "name": "ungrouped_agents",
+          "description": "ungrouped_agents (parquet type: int64)",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "ungrouped_agents"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/event_center_px",
+          "name": "event_center_px",
+          "description": "event_center_px (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "event_center_px"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/event_center_m",
+          "name": "event_center_m",
+          "description": "event_center_m (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "event_center_m"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goal_location_raw",
+          "name": "goal_location_raw",
+          "description": "goal_location_raw (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goal_location_raw"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goals_px",
+          "name": "goals_px",
+          "description": "goals_px (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goals_px"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goals_m",
+          "name": "goals_m",
+          "description": "goals_m (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goals_m"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/initial_state",
+          "name": "initial_state",
+          "description": "initial_state (parquet type: list<element: list<element: double>>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "initial_state"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/groups",
+          "name": "groups",
+          "description": "groups (parquet type: list<element: list<element: int64>>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "groups"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/towards_event",
+          "name": "towards_event",
+          "description": "towards_event (parquet type: bool)",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "towards_event"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/assets",
+          "name": "assets",
+          "description": "assets (parquet type: struct<anchored_obstacles: string, homography: string, obstacle_png: string>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "assets"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/desired_speed_range",
+          "name": "desired_speed_range",
+          "description": "desired_speed_range (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "desired_speed_range"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/collision_rate_early",
+          "name": "collision_rate_early",
+          "description": "collision_rate_early (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "collision_rate_early"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/collision_rate_late",
+          "name": "collision_rate_late",
+          "description": "collision_rate_late (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "collision_rate_late"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/towards_goal",
+          "name": "towards_goal",
+          "description": "towards_goal (parquet type: bool)",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "towards_goal"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/relavance_to_event_range",
+          "name": "relavance_to_event_range",
+          "description": "relavance_to_event_range (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "relavance_to_event_range"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/lingering_fraction_range",
+          "name": "lingering_fraction_range",
+          "description": "lingering_fraction_range (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "lingering_fraction_range"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goal_location_px",
+          "name": "goal_location_px",
+          "description": "goal_location_px (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goal_location_px"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goal_location_m",
+          "name": "goal_location_m",
+          "description": "goal_location_m (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goal_location_m"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "rai:dataLimitations": "The scene descriptions and agent distributions are generated by AI. It may not indicate any real world scenarios.",
+  "rai:dataBiases": "We sampled only public areas with a large crowd of people; We also strategically selected rarer scenarios such as violence and escaping, that are underrepresented in the real world.",
+  "rai:dataUseCases": "This map is used as an input for text-to-trajectory generation. The model could use only the description or if other metadata is needed.",
+  "rai:dataSocialImpact": "As STRIDE-Bench is used for text-to-trajectory generation, it could facilitate in urban planning, venue safety auditing, and emergency preparedness.",
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Scenerio generation",
+      "sc:description": "We adopt Berlonghi's typology of 11 crowd categories~\\cite{berlonghi1995understanding}. For each category, we prompt GPT-5.1 with a structured template that elicits: (i) a natural-language scenario description, (ii) candidate exits and event-center coordinates in the map pixel grid, and (iii) the initial crowd size and spatial distribution. Fields (ii) and (iii) provide a common initialization across models, allowing us to isolate context-conditioned behavior from setup variability. Prompts are provided in Appendix~\\ref{app:scenario-corpus}.\n\\textbf{Filtering and stratified sampling.}\nWe remove scenarios with insufficient agent density, exits or event centers that fall on obstacles, and near-duplicate descriptions. Obstacle conflicts are a known failure mode of polygon approximation, while duplicates are identified by cosine-similarity filtering over sentence embeddings. We also conduct a manual pass to remove residual map-realism failures.\nFrom the filtered pool, we perform \\emph{stratified oversampling} toward safety-critical categories, such as Escaping, Violent, and Dense. These categories are systematically underrepresented in real-world trajectory datasets~\\cite{pellegrini2009you, lerner2007crowds, robicquet2016learning}, since such dynamics cannot be ethically staged. They are also the regimes where context-conditioned generation has the highest applied value, including urban planning, venue safety auditing, and emergency preparedness. For each retained scenario, we prompt GPT-5.2 to generate the behavioral questions, measurements, and expected answers. Prompts are provided in Appendix~\\ref{app:qa-generation}.\n\\textbf{Calibration.}\nTo improve the human alignment of expected answers, we incorporate real-world crowd reference data into our human-trajectory knowledge base, \\emph{TrajFacts}. These references include events such as the Love Parade~\\cite{zhao2020assessing, krausz2012loveparade} and the Itaewon crowd crush~\\cite{hwang2025virtual}. \\emph{TrajFacts} provides reference values that help the model produce more realistic expected-answer ranges. Appendix~\\ref{app:gt-library} gives additional details.\nIn total, STRIDE-Bench contains 936 scenarios, 6,633 questions, and 11,696 measurements. Figure~\\ref{fig:benchmark-stats} shows the heatmap of DMT function usage across crowd categories."
+    }
+  ]
+}

croissant/scenes.croissant.json

@@ -0,0 +1,449 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "repeated": "cr:repeated",
+    "sc": "https://schema.org/",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform"
+  },
+  "@type": "sc:Dataset",
+  "name": "STRIDE-Bench-scenes",
+  "description": "Crowd-simulation scenes for STRIDE-Bench: scenario metadata, initial agent states (position/velocity/goal/radius), group structure, and per-scene expected behavioral ranges. Distributed as a single Parquet table with one row per scene; nested fields whose runtime types vary (e.g. goal_location_*) are stored as JSON-encoded strings.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "license": "https://choosealicense.com/licenses/mit/",
+  "url": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark",
+  "version": "1.0.0",
+  "datePublished": "2026-04-20",
+  "keywords": [
+    "crowd-simulation",
+    "trajectory-evaluation",
+    "pedestrian-dynamics",
+    "benchmark",
+    "behavioral-evaluation"
+  ],
+  "creator": {
+    "@type": "Person",
+    "name": "Anonymous (NeurIPS 2026 D&B submission, double-blind)"
+  },
+  "rai:dataCollectionType": [
+    "Synthetic",
+    "Simulation"
+  ],
+  "rai:personalSensitiveInformation": "None. The dataset contains no personal data, no images of identifiable individuals, and no real pedestrian tracks. All scenes are synthetic.",
+  "rai:hasSyntheticData": true,
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "scenes-parquet",
+      "name": "scenes.parquet",
+      "description": "Parquet table of crowd-simulation scenes: scenario metadata, initial agent states, group structure, goal locations, and expected behavioral ranges. One row per scene.",
+      "contentUrl": "https://huggingface.co/datasets/sweetspot000/tqa-benchmark/resolve/main/scenes.parquet",
+      "encodingFormat": "application/x-parquet",
+      "sha256": "b73edd5a977fc7525330a83ba8a61218953933febdd7de66b27116ff68bd835f",
+      "contentSize": "5641744 B"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "scenes",
+      "name": "scenes",
+      "description": "One record per scene; columns derived from the Parquet schema.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/scene_id",
+          "name": "scene_id",
+          "description": "scene_id (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "scene_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/scene_index",
+          "name": "scene_index",
+          "description": "scene_index (parquet type: int64)",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "scene_index"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/scenario",
+          "name": "scenario",
+          "description": "scenario (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "scenario"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/category",
+          "name": "category",
+          "description": "category (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "category"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/crowd_size_label",
+          "name": "crowd_size_label",
+          "description": "crowd_size_label (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "crowd_size_label"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/crowd_size",
+          "name": "crowd_size",
+          "description": "crowd_size (parquet type: int64)",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "crowd_size"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/ungrouped_agents",
+          "name": "ungrouped_agents",
+          "description": "ungrouped_agents (parquet type: int64)",
+          "dataType": "sc:Integer",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "ungrouped_agents"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/event_center_px",
+          "name": "event_center_px",
+          "description": "event_center_px (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "event_center_px"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/event_center_m",
+          "name": "event_center_m",
+          "description": "event_center_m (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "event_center_m"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goal_location_raw",
+          "name": "goal_location_raw",
+          "description": "goal_location_raw (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goal_location_raw"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goals_px",
+          "name": "goals_px",
+          "description": "goals_px (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goals_px"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goals_m",
+          "name": "goals_m",
+          "description": "goals_m (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goals_m"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/initial_state",
+          "name": "initial_state",
+          "description": "initial_state (parquet type: list<element: list<element: double>>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "initial_state"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/groups",
+          "name": "groups",
+          "description": "groups (parquet type: list<element: list<element: int64>>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "groups"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/towards_event",
+          "name": "towards_event",
+          "description": "towards_event (parquet type: bool)",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "towards_event"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/assets",
+          "name": "assets",
+          "description": "assets (parquet type: struct<anchored_obstacles: string, homography: string, obstacle_png: string>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "assets"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/desired_speed_range",
+          "name": "desired_speed_range",
+          "description": "desired_speed_range (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "desired_speed_range"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/collision_rate_early",
+          "name": "collision_rate_early",
+          "description": "collision_rate_early (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "collision_rate_early"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/collision_rate_late",
+          "name": "collision_rate_late",
+          "description": "collision_rate_late (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "collision_rate_late"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/towards_goal",
+          "name": "towards_goal",
+          "description": "towards_goal (parquet type: bool)",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "towards_goal"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/relavance_to_event_range",
+          "name": "relavance_to_event_range",
+          "description": "relavance_to_event_range (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "relavance_to_event_range"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/lingering_fraction_range",
+          "name": "lingering_fraction_range",
+          "description": "lingering_fraction_range (parquet type: list<element: double>)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "lingering_fraction_range"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goal_location_px",
+          "name": "goal_location_px",
+          "description": "goal_location_px (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goal_location_px"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "scenes/goal_location_m",
+          "name": "goal_location_m",
+          "description": "goal_location_m (parquet type: string)",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "scenes-parquet"
+            },
+            "extract": {
+              "column": "goal_location_m"
+            }
+          }
+        }
+      ]
+    }
+  ]
+}

maps.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a15c3423dba429e9c805c05eb4afb42525efacce7931dd1991a05664cc0c6c4
+size 7735446

scenes.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b73edd5a977fc7525330a83ba8a61218953933febdd7de66b27116ff68bd835f
+size 5641744