Prepare dataset for Hugging Face

.gitattributes

@@ -0,0 +1,2 @@
+metadata.jsonl filter=lfs diff=lfs merge=lfs -text
+images/* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -2,16 +2,105 @@
 
 CUEBench is a neurosymbolic benchmark that emphasizes **contextual entity prediction** in autonomous driving scenes. Unlike traditional detection tasks, CUEBench focuses on reasoning over **unobserved entities** — objects that may be occluded, out-of-frame, or affected by sensor failures.
 
-## Task
+## Dataset Summary
 
-**Input**: A scene ID and a set of `observed_classes` present in the scene
-**Output**: Predict the `target_classes` that were present but unobserved
+- **Modalities**: RGB dashcam imagery + symbolic annotations (provided as metadata)
+- **Primary task**: Predict unobserved `target_classes` given the set of `observed_classes` in a scene
+- **Geography / Scenario**: Urban autonomous driving across diverse traffic densities
+- **License**: CC-BY-4.0 (you may adapt if different licensing is desired)
 
-### Example
+## Dataset Structure
+
+### Data Fields
+| Field | Type | Description |
+| --- | --- | --- |
+| `image_id` | `string` | Unique identifier for each frame (`aligned_id` in the raw metadata).
+| `image_path` | `string` | Relative path to the rendered frame image.
+| `observed_classes` | `list[string]` | Entity classes detected in-frame (cars, cones, pedestrians, etc.).
+| `target_classes` | `list[string]` | Entities inferred to exist but unobserved (occluded, off-frame, sensor failure).
+
+### Splits
+Currently only a **train** split is defined via `metadata.jsonl`. Additional splits can be created before upload if desired (e.g., hold out 10% for validation).
+
+### Label Taxonomy
+Representative classes include: `Car`, `Bus`, `Pedestrian`, `PickupTruck`, `MediumSizedTruck`, `Animal`, `Standing`, `VehicleWithRider`, `ConstructionSign`, `TrafficCone`, and more (~40 classes). Extend this section with the final taxonomy before publication if you want exhaustive documentation.
+
+## Example Record
 ```json
 {
   "image_id": "00003.00019",
   "observed_classes": ["Car", "Bus", "Pedestrian"],
-  "target_classes": ["PickupTruck"]
+  "target_classes": ["PickupTruck"],
+  "image_path": "images/00003.00019.png"
 }
+```
+
+## Usage
+
+### Loading with `datasets`
+```python
+from datasets import load_dataset
+
+dataset = load_dataset(
+  path="ishwarbb/cuebench",
+    split="train"
+)
+```
+
+### Working From Source
+```python
+from datasets import load_dataset
+
+dataset = load_dataset(path="cuebench", data_files="metadata.jsonl", split="train")
+```
+
+## Metrics
+
+`metric.py` defines **Mean Reciprocal Rank**, **Hits@K (1/3/5/10)**, and **Coverage@K (1/3/5/10)** over the predicted class rankings. When publishing to the Hugging Face Metrics Hub, expose the `compute(predictions, references)` signature so leaderboard integrations can consume it.
+
+## Licensing
+
+The dataset is currently tagged as **CC-BY-4.0** in `cuebench.py`. Update this section if you select a different license.
+
+## Citation
+
+```
+@misc{cuebench2025,
+  title  = {CUEBench: Contextual Unobserved Entity Benchmark},
+  author = {CUEBench Authors},
+  year   = {2025}
+}
+```
+
+## Hugging Face Upload Checklist
+
+1. Install tools: `pip install datasets huggingface_hub` and run `huggingface-cli login`.
+2. Create the dataset repo: `huggingface-cli repo create cuebench --type dataset` (or via UI).
+3. Ensure directory layout:
+   ```
+   cuebench/
+     README.md
+     cuebench.py
+     metadata.jsonl
+     metric.py           # optional metric script
+     images/...          # optional or host separately
+   ```
+4. Initialize Git + LFS:
+   ```bash
+   cd cuebench
+   git init
+   git lfs install
+   git lfs track "*.jsonl" "images/*"
+  git remote add origin https://huggingface.co/datasets/ishwarbb/cuebench
+   git add .
+   git commit -m "Initial CUEBench dataset"
+   git push origin main
+   ```
+5. Validate locally before pushing updates (optional but recommended):
+   - `datasets-cli test ./cuebench.py --all_configs`
+   - `python -m datasets.prepare_module ./cuebench.py`
+6. On the Hub page, trigger the dataset preview to ensure the loader runs.
+7. (Optional) Publish the metric under `metrics/cuebench-metric` following the Metrics Hub template and link it from the dataset card.
+
+Update these steps with any organization-specific tooling you use.
 

cuebench.py

@@ -1,7 +1,27 @@
 import json
-from datasets import DatasetInfo, GeneratorBasedBuilder, SplitGenerator, Split, Value, Features, Sequence
+from datasets import (
+    BuilderConfig,
+    DatasetInfo,
+    Features,
+    GeneratorBasedBuilder,
+    Sequence,
+    Split,
+    SplitGenerator,
+    Value,
+    Version,
+)
 
 class CUEBench(GeneratorBasedBuilder):
+    VERSION = Version("1.0.0")
+    BUILDER_CONFIGS = [
+        BuilderConfig(
+            name="default",
+            version=VERSION,
+            description="Contextual Unobserved Entity Benchmark leveraging autonomous driving scenes.",
+        )
+    ]
+    DEFAULT_CONFIG_NAME = "default"
+
     def _info(self):
         return DatasetInfo(
             description="CUEBench: Contextual Entity Prediction for Occluded or Unobserved Entities in Autonomous Driving.",
@@ -11,24 +31,30 @@ class CUEBench(GeneratorBasedBuilder):
                 "target_classes": Sequence(Value("string")),
                 "image_path": Value("string")
             }),
-            citation="",
-            homepage=""
+            citation="@misc{cuebench2025, title={CUEBench: Contextual Unobserved Entity Benchmark}, year={2025}, author={CUEBench Authors}}",
+            homepage="https://huggingface.co/datasets/ishwarbb/cuebench",
+            license="CC-BY-4.0",
         )
 
     def _split_generators(self, dl_manager):
-        data_files = self.config.data_files
-        filepath = dl_manager.download_and_extract(data_files["train"] if isinstance(data_files, dict) else data_files)
+        data_files = self.config.data_files or {"train": "metadata.jsonl"}
+        train_files = data_files["train"] if isinstance(data_files, dict) else data_files
+        filepath = dl_manager.download_and_extract(train_files)
+        if isinstance(filepath, list):
+            filepath = filepath[0]
         return [SplitGenerator(name=Split.TRAIN, gen_kwargs={"filepath": filepath})]
 
     def _generate_examples(self, filepath):
-        print("f = ", filepath)
         if isinstance(filepath, list):
             filepath = filepath[0]
         with open(filepath, "r", encoding="utf-8") as f:
             for idx, line in enumerate(f):
                 example = json.loads(line)
+                image_id = example.get("aligned_id") or example.get("image_id")
+                if image_id is None:
+                    raise ValueError(f"Missing image identifier for example at line {idx}.")
                 yield idx, {
-                    "image_id": example["aligned_id"],  # Ensure this key exists in your JSONL
+                    "image_id": image_id,
                     "image_path": example["image_path"],
                     "observed_classes": example["detected_classes"],  # Already a list
                     "target_classes": example["target_classes"],

requirements.txt

@@ -0,0 +1,5 @@
+datasets==2.14.6
+pyarrow==11.0.0
+numpy==1.26.4
+pandas==2.3.3
+huggingface_hub==0.36.0