Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
task_categories:
|
| 4 |
+
- image-to-text
|
| 5 |
+
- robotics
|
| 6 |
+
tags:
|
| 7 |
+
- autonomous-driving
|
| 8 |
+
- nuScenes
|
| 9 |
+
- nuReasoning
|
| 10 |
+
- trajectory-prediction
|
| 11 |
+
- world-model
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# WAM-data (GeoWAM Stage-2 extracted data)
|
| 15 |
+
|
| 16 |
+
从 `qixuewei/nuReasoning` 抽取的 GeoWAM / MMaDA Stage-2 训练数据的**元数据**。
|
| 17 |
+
|
| 18 |
+
## 仓库内容
|
| 19 |
+
|
| 20 |
+
| 文件 | 内容 |
|
| 21 |
+
|---|---|
|
| 22 |
+
| `WAM_data_meta.tar.gz` | `manifest.jsonl` + `ego_state.jsonl` + `done_clips.txt` + `ego_done_clips.txt` |
|
| 23 |
+
|
| 24 |
+
- **manifest.jsonl**（7785 行）：每个决策关键帧一条样本，字段
|
| 25 |
+
`clip, cur(6视图图片名), fut, prompt, reasoning, spatial, geom, command`
|
| 26 |
+
- **ego_state.jsonl**（7920 行）：每帧 ego 运动学状态 + 5s 未来轨迹
|
| 27 |
+
`clip, fidx, speed, vx, vy, ax, ay, px, py, yaw, traj_future[[x,y,yaw]...]`
|
| 28 |
+
|
| 29 |
+
## 图片未包含（55GB，5.5万张）
|
| 30 |
+
|
| 31 |
+
图片体量太大未上传。`manifest.jsonl` 里记录了每张图片的文件名，可用项目里的
|
| 32 |
+
`training/extract_nureasoning.py` 从原始 `qixuewei/nuReasoning` 重新抽取（断点续传）：
|
| 33 |
+
|
| 34 |
+
```bash
|
| 35 |
+
python -m training.extract_nureasoning --n_clips 0 --workers 24
|
| 36 |
+
```
|
| 37 |
+
|
| 38 |
+
抽出的图片落在 `images/<clip>_<fidx>_cur_<view>.jpg` 和 `<clip>_<fidx>_fut.jpg`，
|
| 39 |
+
与 manifest 中的文件名一一对应。
|
| 40 |
+
|
| 41 |
+
## 配套代码 / 模型
|
| 42 |
+
|
| 43 |
+
- 代码 + 环境 + 对话历史 + Stage-1 backbone：`ziyuz/WAM`（model repo）
|