futurefantasy commited on
Commit
8aac3db
·
verified ·
1 Parent(s): 13c644b

Delete docs/vlac2_public_release.md

Browse files
Files changed (1) hide show
  1. docs/vlac2_public_release.md +0 -121
docs/vlac2_public_release.md DELETED
@@ -1,121 +0,0 @@
1
- # VLAC2 Public Release
2
-
3
- 这是一个最小使用说明,主要说明拿到发布包后怎么放数据、怎么抽帧、以及后续该用哪个 benchmark JSON。
4
-
5
- ## 发布包包含什么
6
-
7
- ```text
8
- vlac2_release/
9
- ├── benchmark_json/
10
- ├── vlac2_release_data/
11
- │ ├── vlac2_release_data_part*.tar.zst
12
- │ └── vlac2_release_data_archives_manifest.json
13
- ├── scripts/
14
- └── docs/
15
- ```
16
-
17
- - `benchmark_json/`
18
- - 公开提供的 benchmark 文件。
19
- - 包含 `train/` 和 4 个 `test_*` bucket。
20
- - `vlac2_release_data/`
21
- - 放原始数据集分片归档。
22
- - 解压全部归档后,会得到 `ARX-data/`、`droid_lerobot/`、`libero/` 等原始数据目录。
23
- - `scripts/extract_vlac2_release_raw_archives.sh`
24
- - 先把 `vlac2_release_data/` 下所有 `vlac2_release_data_part*.tar.zst` 解压到 `vlac2_release_data/`。
25
- - `scripts/extract_vlac2_release_frames.py`
26
- - 原始数据解压完成后,再运行这个脚本做抽帧。
27
- - 如果不手动指定 `--output-root`,脚本会自动把图片写到 `<release-root>/_extracted_frames/`,并在首次运行时创建该目录。
28
-
29
- ## 基础使用
30
-
31
- ### 1. 解压原始数据
32
-
33
- 先把 `vlac2_release_data/` 下的所有归档解压到:
34
-
35
- ```text
36
- <release-root>/vlac2_release_data/
37
- ```
38
-
39
- 最简单的方式是直接运行:
40
-
41
- ```bash
42
- bash scripts/extract_vlac2_release_raw_archives.sh /path/to/vlac2_release/vlac2_release_data
43
- ```
44
-
45
- 如果你想手动解压,也可以逐个运行:
46
-
47
- ```bash
48
- for archive in /path/to/vlac2_release/vlac2_release_data/vlac2_release_data_part*.tar.zst; do
49
- tar --zstd -xf "${archive}" -C /path/to/vlac2_release/vlac2_release_data
50
- done
51
- ```
52
-
53
- 解压完成后,目录结构应当类似于:
54
-
55
- ```text
56
- <release-root>/vlac2_release_data/ARX-data/...
57
- <release-root>/vlac2_release_data/droid_lerobot/...
58
- <release-root>/vlac2_release_data/libero/...
59
- <release-root>/vlac2_release_data/libero_zty50/...
60
- <release-root>/vlac2_release_data/VLABench_5/...
61
- ```
62
-
63
- ### 2. 运行抽帧
64
-
65
- 在 release 根目录下运行:
66
-
67
- ```bash
68
- python scripts/extract_vlac2_release_frames.py --release-root /path/to/vlac2_release
69
- ```
70
-
71
- 默认行为:
72
-
73
- - 自动读取 `benchmark_json/` 里的 benchmark 文件
74
- - 自动找到对应原始视频
75
- - 如果未指定 `--output-root`,把抽出的帧写到 `_extracted_frames/...`
76
- - 图片命名格式为:
77
- - `{frame_idx}-{num_frames}.jpg`
78
-
79
- 如果只想抽 benchmark 主视角,可以运行:
80
-
81
- ```bash
82
- python scripts/extract_vlac2_release_frames.py \
83
- --release-root /path/to/vlac2_release \
84
- --view-mode main-only
85
- ```
86
-
87
- 如果想强制重抽已经存在的帧,可以运行:
88
-
89
- ```bash
90
- python scripts/extract_vlac2_release_frames.py \
91
- --release-root /path/to/vlac2_release \
92
- --overwrite-existing
93
- ```
94
-
95
- ## 后续使用哪个文件
96
-
97
- 后续测评只使用:
98
-
99
- - `benchmark_json/.../video_progress_benchmark_file.json`
100
-
101
- 不需要用户自己重建 benchmark JSON。
102
-
103
- ## 路径说明
104
-
105
- `video_progress_benchmark_file.json` 里的图片路径使用占位前缀:
106
-
107
- - `__VLAC2_RELEASE_ROOT__/_extracted_frames/...`
108
-
109
- 实际使用时,把它替换成你本地 release 根目录的绝对路径即可。例如:
110
-
111
- - `__VLAC2_RELEASE_ROOT__/_extracted_frames/...`
112
- - `/abs/path/to/vlac2_release/_extracted_frames/...`
113
-
114
- ## 依赖
115
-
116
- 抽帧脚本至少需要一个解码后端:
117
-
118
- - `opencv-python`
119
- - 或 `imageio`
120
-
121
- 如果两者都没有安装,脚本会直接报错退出。