MonsterDie commited on
Commit
e83c9cc
·
verified ·
1 Parent(s): 383eafb

Translate dataset README to English and link pipeline

Browse files
Files changed (1) hide show
  1. README.md +35 -28
README.md CHANGED
@@ -4,47 +4,54 @@ license: mit
4
 
5
  # SpatialSemanticDataset
6
 
7
- SpatialSemanticDataset 是一个面向第一人称步行视频的空间语义数据集。它按场景组织原视频的时间区间、带时间戳的相机位姿和英文语义标注,可用于研究视觉导航、相机轨迹与语言对齐,以及场景观察和动作意图描述。
8
 
9
- 当前版本包含 **9 个来源频道、3,188 个视频和 12,079 个场景**。仓库提供索引与标注文件,原始视频可通过 `video_id` 定位,视频文件未随仓库分发。
10
 
11
- ## 文件结构
12
 
13
- | 路径 | 内容 |
 
 
14
  | --- | --- |
15
- | `metadata.csv` | 场景索引,每行对应一个场景 |
16
- | `poses/<channel>/<video_id>/<scene_id>.csv` | 相机位置、朝向及位姿状态 |
17
- | `semantics/<channel>/<video_id>/<scene_id>.txt` | 带时间戳的英文观察、推断与动作描述 |
18
 
19
- 以 `(channel, video_id, scene_id)` 作为联合键查找同一场景的两个标注文件。`scene_id` 是视频内的场景编号,例如 `scene_0000`,在不同视频中可以重复。
20
 
21
- ## 数据字段
22
 
23
- ### 场景索引
24
 
25
- - `channel` 和 `video_id` 分别标识来源频道与 YouTube 视频。
26
- - `scene_id` 标识视频内的场景。
27
- - `start_second` 和 `end_second` 记录场景在原视频中的起止秒数。
 
 
 
28
 
29
- ### 相机位姿
30
 
31
- | 字段 | 含义 |
32
  | --- | --- |
33
- | `source_t` | 原视频时间,单位为秒 |
34
- | `source_frame` | 原视频帧编号 |
35
- | `status` | 位姿状态。样本中的 `valid` 表示位姿可用,`bad` 行的位姿字段可为 `NaN` |
36
- | `x`, `y`, `z` | 相机位置分量 |
37
- | `qx`, `qy`, `qz`, `qw` | 相机朝向的四元数分量,标量分量 `qw` 在末尾 |
 
 
38
 
39
- 仓库未说明位置单位、坐标轴约定或跨场景坐标系关系。使用位姿计算物理距离或拼接轨迹前,需要另行确认这些约定。
40
 
41
- ### 语义标注
42
 
43
- 文本按原视频时间戳 `[HH:MM:SS]` 分块。每块包含 `Observation`(可见场景描述)、`Latent Thought`(以第一人称表述的情境或意图推断)和 `Action`(动作或行动计划描述)。其中的想法和动作应按文本标注理解。位姿与语义标注的时间点可以不同,可先按场景关联,再按原视频时间对齐。
44
 
45
- ## 读取一个场景
46
 
47
- 该仓库是私有的。安装 `huggingface_hub`,通过 `hf auth login` 登录有读取权限的账户,或设置该账户的 `HF_TOKEN` 环境变量。以下示例只下载索引与首个场景的两个文件。
48
 
49
  ```python
50
  import csv
@@ -74,8 +81,8 @@ print(f"Valid poses: {len(valid_poses)}")
74
  print(annotations.split("\n\n", 1)[0])
75
  ```
76
 
77
- 仓库未提供预定义的数据划分。如需构建训练集与评估集,可按 `(channel, video_id)` 分组,避免同一视频的不同场景跨集合。
78
 
79
- ## 许可
80
 
81
- 数据集卡沿用仓库声明的 MIT 许可。原始视频的使用条件由其来源决定。
 
4
 
5
  # SpatialSemanticDataset
6
 
7
+ SpatialSemanticDataset is a dataset of first-person walking videos annotated with camera poses and English semantic descriptions. It organizes annotations by scene and supports research on visual navigation, camera trajectory and language alignment, and descriptions of observations and intended actions.
8
 
9
+ The dataset covers **12,079 scenes from 3,188 videos across 9 source channels**. This repository contains the scene index and annotations. The source videos are hosted separately and can be located by their YouTube video IDs.
10
 
11
+ The data processing pipeline is maintained in a separate repository. [SpatialSemanticPipeline](https://github.com/monsterdie12138/SpatialSemanticPipeline)
12
 
13
+ ## Repository layout
14
+
15
+ | Path | Contents |
16
  | --- | --- |
17
+ | `metadata.csv` | Scene index, with one row per scene |
18
+ | `poses/<channel>/<video_id>/<scene_id>.csv` | Timestamped camera positions, orientations, and pose status |
19
+ | `semantics/<channel>/<video_id>/<scene_id>.txt` | Timestamped English observations, inferences, and action descriptions |
20
 
21
+ Use `(channel, video_id, scene_id)` to join the index with the two annotation files. Scene IDs such as `scene_0000` are local to each video.
22
 
23
+ ## Data fields
24
 
25
+ ### Scene index
26
 
27
+ | Field | Meaning |
28
+ | --- | --- |
29
+ | `channel` | Source channel |
30
+ | `video_id` | YouTube video ID |
31
+ | `scene_id` | Scene identifier within the source video |
32
+ | `start_second`, `end_second` | Scene start and end times in the source video, in seconds |
33
 
34
+ ### Camera poses
35
 
36
+ | Field | Meaning |
37
  | --- | --- |
38
+ | `source_t` | Timestamp in the source video, in seconds |
39
+ | `source_frame` | Frame index in the source video |
40
+ | `status` | Pose status. Sample files contain `valid` rows with pose values and `bad` rows whose pose fields can be `NaN`. |
41
+ | `x`, `y`, `z` | Camera position components |
42
+ | `qx`, `qy`, `qz`, `qw` | Camera orientation quaternion, with the scalar component `qw` last |
43
+
44
+ The dataset files leave the position units, axis conventions, and coordinate relationships between scenes unspecified. Confirm these conventions before calculating physical distances or combining trajectories across scenes.
45
 
46
+ ### Semantic annotations
47
 
48
+ Annotations are grouped by source-video timestamp in `[HH:MM:SS]` format. Each block contains an `Observation` describing what is visible, a `Latent Thought` expressing an inferred situation or intent in the first person, and an `Action` describing an action or plan.
49
 
50
+ Pose samples and semantic annotations can occur at different times. Join them by scene, then align them on the source-video timeline.
51
 
52
+ ## Load a scene
53
 
54
+ This repository is private. Install `huggingface_hub` and use `hf auth login` or set the `HF_TOKEN` environment variable with a token that has read access. This example downloads the index and the two annotation files for its first scene.
55
 
56
  ```python
57
  import csv
 
81
  print(annotations.split("\n\n", 1)[0])
82
  ```
83
 
84
+ The dataset ships without predefined splits. When creating training and evaluation splits, group scenes by `(channel, video_id)` so that scenes from the same source video stay in the same split.
85
 
86
+ ## License
87
 
88
+ The dataset card declares the MIT license. Source videos are subject to their own terms.