| --- |
| license: cc-by-nc-sa-4.0 |
| task_categories: |
| - object-detection |
| tags: |
| - wildfire |
| - smoke |
| - fire-detection |
| - fire-origin |
| pretty_name: HPWREN Fire-Origin Points (dev eval) |
| --- |
| |
| # HPWREN Fire-Origin Points — development eval set |
|
|
| Human-placed **fire-origin points** (the base of the smoke column — the point that feeds |
| azimuth / geolocation) on **confirmed-wildfire** frames from HPWREN fixed cameras. |
| 31 frames spanning ~31 distinct fires. |
|
|
| ## ⚠️ Provenance & license |
| The **images** are HPWREN wildfire-camera frames annotated with smoke bounding boxes by |
| **[AI For Mankind](https://github.com/aiformankind/wildfire-smoke-dataset)**, released |
| under **CC BY-NC-SA 4.0**. This dataset **redistributes a curated subset of those images |
| under the same license** — *attribution, non-commercial, share-alike*. Please credit |
| **AI For Mankind** and **HPWREN**. The **fire-origin annotations** are the contribution |
| added here (also CC BY-NC-SA 4.0). |
|
|
| ## Columns |
| `image`, `annotated` (overlay: GT box + origin crosshair + uncertainty region), |
| `x_origin`/`y_origin` (normalized origin), `x_min…y_max` (2D uncertainty region), |
| `base_occluded`, `confidence` (clear/moderate/guess), `outside_box` (+`outside_dist`, |
| derived), `box_*`, `plume_size`, `scene`, `camera`, `annotator`. |
|
|
| ## Curation |
| 744 AI For Mankind frames are FIgLib ignition *sequences* (many near-identical frames |
| per fire). We deduplicated to unique fires (camera + date-region hashing), then |
| hand-curated — best frame per fire, dropping non-fires — to ~31 confirmed fires. |
|
|
| ## Intended use |
| A **small dev/eval set** for fire-origin estimation methods (bbox-center baseline, |
| classical CV, SAM). **Not a benchmark** — small N, single region (Southern California). |
| Non-commercial research only. |
|
|
