Use numeric Nexar video ids; document nexar-ai/nexar_collision_prediction as video source

README.md

@@ -50,7 +50,9 @@ dataset_info:
 
 **Causal Accident Video and Incident Analysis Repository** — Nexar validation/test annotations.
 
-This Hub dataset mirrors the public val/test split from [github.com/nec-labs-ma/CAViAR](https://github.com/nec-labs-ma/CAViAR): **749 videos**, **7,407 QA pairs**. **Videos are not included.** Obtain Nexar clips under the original Nexar license; `video_path` is a filename only (e.g. `nexar_00284.mp4`).
+This Hub dataset mirrors the public val/test split from [github.com/nec-labs-ma/CAViAR](https://github.com/nec-labs-ma/CAViAR): **749 videos**, **7,407 QA pairs**. **Videos are not included.**
+
+`video_path` is the **numeric Nexar clip id** (e.g. `00284`). Source videos: [nexar-ai/nexar_collision_prediction](https://huggingface.co/datasets/nexar-ai/nexar_collision_prediction) (`{id}.mp4` under `train/` / `test-public/` / `test-private/`).
 
 Paper: *CAViAR: A Causal Video Dataset for Fine-Grained Accident Reasoning in Real-World Scenarios* ([arXiv:2608.19380](https://arxiv.org/abs/2608.19380)).
 
@@ -60,17 +62,17 @@ Paper: *CAViAR: A Causal Video Dataset for Fine-Grained Accident Reasoning in Re
 from datasets import load_dataset
 
 ds = load_dataset("sparshgarg57/CAViAR", split="test")
-print(ds[0]["video_path"], len(ds[0]["qa_pairs"]))
+print(ds[0]["video_path"], len(ds[0]["qa_pairs"]))  # e.g. "00776", 10
 ```
 
-Canonical JSON (same as the GitHub file): [`data/test.json`](data/test.json).
+Canonical JSON (same as GitHub): [`data/test.json`](data/test.json).
 
 ## Split
 
 | Split | Source | Role | Released here |
 |-------|--------|------|----------------|
 | Train | [CCD](https://github.com/Cogito2012/CarCrashDataset) | Fine-tuning | No |
-| Val / test | [Nexar](https://www.nexar.com/) | Held-out evaluation | Yes (annotations only) |
+| Val / test | [nexar-ai/nexar_collision_prediction](https://huggingface.co/datasets/nexar-ai/nexar_collision_prediction) | Held-out evaluation | Yes (annotations only) |
 
 ## Tasks
 
@@ -89,25 +91,28 @@ MCQ items include `choices`, `correct_answer`, and `correct_index`. Open-ended i
 
 ## Videos
 
-Place Nexar mp4 files so names match `video_path`:
+Download from [nexar-ai/nexar_collision_prediction](https://huggingface.co/datasets/nexar-ai/nexar_collision_prediction) (Nexar Open Data License). Join on id:
+
+| CAViAR `video_path` | Nexar file |
+|---------------------|------------|
+| `00284` | `00284.mp4` |
 
-```text
-nexar_00001.mp4
-...
-nexar_00284.mp4
+```bash
+huggingface-cli download nexar-ai/nexar_collision_prediction --repo-type dataset --local-dir /path/to/nexar_collision_prediction
+export CAVIAR_VIDEO_ROOT=/path/to/nexar_collision_prediction
 ```
 
-Code, eval scripts, and ontology: [CAViAR GitHub](https://github.com/nec-labs-ma/CAViAR).
+Eval scripts search `train/`, `test-public/`, and `test-private/` for `{id}.mp4`. Code: [CAViAR GitHub](https://github.com/nec-labs-ma/CAViAR).
 
-Example clip (`nexar_00284`, T-bone):
+Example clip (`00284`, T-bone):
 
-![Preview of nexar_00284.mp4](https://github.com/nec-labs-ma/CAViAR/raw/main/examples/nexar_00284.gif)
+![Preview of Nexar video 00284](https://github.com/nec-labs-ma/CAViAR/raw/main/examples/nexar_00284.gif)
 
-Source video from Nexar (Moura et al., 2025). Short collision-window preview only; the Nexar corpus is not redistributed.
+Short collision-window preview only; the Nexar corpus is not redistributed here.
 
 ## License and ethics
 
-Annotations are for academic research on accident understanding and rule-relevant multimodal reasoning. Raw Nexar (and CCD) videos remain under their original licenses.
+Annotations are for academic research on accident understanding and rule-relevant multimodal reasoning. Raw Nexar videos remain under the [Nexar Open Data License](https://huggingface.co/datasets/nexar-ai/nexar_collision_prediction); CCD videos remain under their original license.
 
 Labels are *apparent* responsibility cues from video evidence. They are **not** legal determinations of liability and must not be used for adjudication, insurance, enforcement, or decisions about identifiable individuals.
 
@@ -122,4 +127,16 @@ Labels are *apparent* responsibility cues from video evidence. They are **not**
 }
 ```
 
-Please also cite **Nexar** when using the val/test videos (Moura et al., 2025).
+Please also cite **Nexar** when using the val/test videos:
+
+```bibtex
+@misc{nexar2025dashcamcollisionprediction,
+  title={Nexar Dashcam Collision Prediction Dataset and Challenge},
+  author={Daniel C. Moura and Shizhan Zhu and Orly Zvitia},
+  year={2025},
+  eprint={2503.03848},
+  archivePrefix={arXiv},
+  primaryClass={cs.CV},
+  url={https://arxiv.org/abs/2503.03848}
+}
+```

data/test.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0b440473b7a99df0d72bec3c4cafceedc5dc1ae3b7476c98730838e28dbacea
-size 145727
+oid sha256:df723dec5a4d7ff28a53d2b2cf0a415f25c97deef463980dfe0a004cb12d42b2
+size 144965