Nurec_eval / README.md
luuuulinnnn's picture
Add 59 OOD-longtail USDZ scenes with HD-map artifacts injected
dab81f5 verified
---
license: cc-by-4.0
task_categories:
- robotics
tags:
- autonomous-driving
- nurec
- closed-loop-eval
- alpasim
- alpamayo
size_categories:
- n<1K
---
# Nurec_eval — NRE Closed-Loop Eval Set with HD Map
**59 OOD-longtail scenes** for AlpaSim closed-loop evaluation, exported from NVIDIA NRE 26.02 and **augmented with HD-map artifacts** (lane graph, road boundaries, crosswalks) so models can be evaluated with full map context — not just geometry.
## What's inside each `pai_<uuid>.usdz`
| File | Purpose |
|---|---|
| `default.usda` | USDZ entry point |
| `checkpoint.ckpt` | Neural reconstruction (NRE-26.02 gsplat) weights |
| `parsed_config.yaml` | NRE training config |
| `data_info.json` | Sequence metadata |
| `datasource_summary.json` | Camera/lidar rig info |
| `mesh.ply` / `mesh.usd` | Poisson reconstructed scene mesh |
| `ground_mesh.ply` | Ground mesh (used for offroad / collision_with_ground) |
| `rig_trajectories.json` / `.usda` | Recorded ego trajectory (route source) |
| `sequence_tracks.json` / `.usda` | Other actors' recorded trajectories |
| **`map_data/cf_crosswalks.parquet`** | **HD-map: crosswalk polygons** |
| **`map_data/cf_lane_topology_node.parquet`** | **HD-map: lane topology** |
| **`map_data/cf_road_boundary.parquet`** | **HD-map: road boundaries** |
| **`map_data/dw_lane.parquet`** | **HD-map: drivable lanes** |
| **`map_data/lane_chunk.parquet`** | **HD-map: lane geometry** |
| **`map_data/lane_rail.parquet`** | **HD-map: lane center rails** |
The 6 `map_data/*.parquet` files were injected via [inject_map.py](https://huggingface.co/datasets/luuuulinnnn/Nurec_eval/blob/main/inject_map.py) using NRE-internal HD-map data.
## Scene composition — 59 scenes from 8 OOD-longtail buckets
| Category | Scenes |
|---|---:|
| Animals / Birds / Roadkill | 7 |
| Complex Intersection Interaction | 7 |
| Cyclists & Micromobility Complex | 8 |
| Emergency Incident Scene | 7 |
| Pedestrian Density / Close Proximity | 8 |
| Road Debris / Safety Traces | 5 |
| Special / Uncommon Vehicle Behavior | 8 |
| Work Zones / Temp Traffic Control | 8 |
| ⚠️ Broken (route folds back) | 1 |
| **Total usable** | **58 × 2 rollouts = 116** |
## Usage with AlpaSim
```bash
# Mount this dataset in the AlpaSim wizard
alpasim_wizard \
deploy=local topology=1gpu driver=<your_model> \
+physics=disabled +vehicle=custom \
scenes.path=/path/to/Nurec_eval \
scenes.scenes_csv=/path/to/Nurec_eval/sim_scenes.csv \
scenes.suites_csv=/path/to/Nurec_eval/sim_suites.csv \
...
```
Note: `physics=disabled` is still required because these are recon-only USDZs (no vehicle-physics config). The HD map is consumed by driver models for lane-aware decision making and by metrics like `wrong_lane` / `offroad-by-lane`.
## What was missing before
Without `map_data/*.parquet`, driver models had **no map context** during closed-loop sim, and lane-aware metrics (`wrong_lane`) couldn't be computed. This dataset adds them back.
## Source
- Scenes exported from NRE 26.02 (`26.2.158-2b8da993`)
- Re-packed with HD map artifacts from `ood_closeloop_map` (79 scenes available, 59 used here)
- Compatible with [NVlabs/alpasim](https://github.com/NVlabs/alpasim)
- Companion datasets: [luuuulinnnn/OOD17_longtail_usdz](https://huggingface.co/datasets/luuuulinnnn/OOD17_longtail_usdz), [luuuulinnnn/OOD64-NuRec-USDZ-with-ground](https://huggingface.co/datasets/luuuulinnnn/OOD64-NuRec-USDZ-with-ground)