Datasets:
Formats:
parquet
Languages:
English
Size:
1K - 10K
ArXiv:
Tags:
embodied-ai
embodied-navigation
urban-airspace
drone-navigation
multimodal-reasoning
spatial-reasoning
License:
Update dataset card for image archive release
Browse filesAdd concise documentation for the released trajectory-aligned image archives under images/, including the five ZIP ranges and the sample_index alignment note.
README.md
CHANGED
|
@@ -43,6 +43,7 @@ The dataset is intended for evaluating embodied navigation, spatial reasoning, a
|
|
| 43 |
| :-- | :-- |
|
| 44 |
| `navi_data.pkl` | Canonical PKL file for evaluation. |
|
| 45 |
| `viewer-00000-of-00001.parquet` | Parquet representation for the Hugging Face Dataset Viewer table. |
|
|
|
|
| 46 |
|
| 47 |
## Data Fields
|
| 48 |
|
|
@@ -50,7 +51,7 @@ The canonical PKL file stores a list of Python dictionaries. Each sample contain
|
|
| 50 |
|
| 51 |
| Field | Type | Description |
|
| 52 |
| :-- | :-- | :-- |
|
| 53 |
-
| `sample_index` | `int` |
|
| 54 |
| `start_pos` | `float[3]` | Initial drone world position `(x, y, z)`. |
|
| 55 |
| `start_rot` | `float[3]` | Initial drone orientation `(roll, pitch, yaw)` in radians. |
|
| 56 |
| `start_ang` | `float` | Initial camera gimbal angle in degrees. |
|
|
@@ -59,7 +60,23 @@ The canonical PKL file stores a list of Python dictionaries. Each sample contain
|
|
| 59 |
| `gt_traj` | `float[N,3]` | Ground-truth trajectory points. |
|
| 60 |
| `gt_traj_len` | `float` | Ground-truth trajectory length. |
|
| 61 |
|
| 62 |
-
The Parquet table includes the same structured fields and additional convenience columns such as `sample_index`, `start_x`, `start_y`, `start_z`, `target_x`, `target_y`, `target_z`, and `gt_traj_num_points`. The
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
|
| 64 |
## Usage
|
| 65 |
|
|
|
|
| 43 |
| :-- | :-- |
|
| 44 |
| `navi_data.pkl` | Canonical PKL file for evaluation. |
|
| 45 |
| `viewer-00000-of-00001.parquet` | Parquet representation for the Hugging Face Dataset Viewer table. |
|
| 46 |
+
| `images/` | Trajectory-aligned image release, distributed as five ZIP archives plus a manifest file. |
|
| 47 |
|
| 48 |
## Data Fields
|
| 49 |
|
|
|
|
| 51 |
|
| 52 |
| Field | Type | Description |
|
| 53 |
| :-- | :-- | :-- |
|
| 54 |
+
| `sample_index` | `int` | Sample index used for viewer browsing and image archive alignment. |
|
| 55 |
| `start_pos` | `float[3]` | Initial drone world position `(x, y, z)`. |
|
| 56 |
| `start_rot` | `float[3]` | Initial drone orientation `(roll, pitch, yaw)` in radians. |
|
| 57 |
| `start_ang` | `float` | Initial camera gimbal angle in degrees. |
|
|
|
|
| 60 |
| `gt_traj` | `float[N,3]` | Ground-truth trajectory points. |
|
| 61 |
| `gt_traj_len` | `float` | Ground-truth trajectory length. |
|
| 62 |
|
| 63 |
+
The Parquet table includes the same structured fields and additional convenience columns such as `sample_index`, `start_x`, `start_y`, `start_z`, `target_x`, `target_y`, `target_z`, and `gt_traj_num_points`. The Parquet file is provided for browsing and visualization in the Hugging Face Dataset Viewer.
|
| 64 |
+
|
| 65 |
+
## Trajectory-Aligned Images
|
| 66 |
+
|
| 67 |
+
Trajectory-aligned image archives are available under [`images/`](https://huggingface.co/datasets/EmbodiedCity/EmbodiedNav-Bench/tree/main/images).
|
| 68 |
+
|
| 69 |
+
This release is about 56.7 GB and is distributed as five ZIP archives together with `merged_upload_images_zip_manifest.json`.
|
| 70 |
+
|
| 71 |
+
After extraction, folders `0-5036` correspond directly to the `sample_index` field in `navi_data.pkl` and the viewer table.
|
| 72 |
+
|
| 73 |
+
| Archive | Sample index range |
|
| 74 |
+
| :-- | :-- |
|
| 75 |
+
| `merged_upload_images_part01_0000-1007.zip` | `0-1007` |
|
| 76 |
+
| `merged_upload_images_part02_1008-2015.zip` | `1008-2015` |
|
| 77 |
+
| `merged_upload_images_part03_2016-3022.zip` | `2016-3022` |
|
| 78 |
+
| `merged_upload_images_part04_3023-4029.zip` | `3023-4029` |
|
| 79 |
+
| `merged_upload_images_part05_4030-5036.zip` | `4030-5036` |
|
| 80 |
|
| 81 |
## Usage
|
| 82 |
|