video-explorer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Video Explorer

Analyze video content through hierarchical frame extraction. Start wide, identify interesting regions, zoom in.

通过分层帧提取分析视频内容。先全局概览，识别感兴趣区域，再局部放大。

Workflow

工作流程

1. Overview First

1. 先全局概览

Extract quick thumbnails to see the video timeline:

bash

./skills/video-explorer/scripts/videx overview <video>

This creates small frames (320px) at 10-second intervals in

./videx-out/<name>/overview/

Read all overview frames to understand video structure:

bash

ls ./videx-out/<name>/overview/

Then use the Read tool on the jpg files to see them.

提取快速缩略图以查看视频时间线：

bash

./skills/video-explorer/scripts/videx overview <video>

这会在

./videx-out/<name>/overview/

目录下生成间隔10秒的小尺寸帧（320像素）。

查看所有概览帧以了解视频结构：

bash

ls ./videx-out/<name>/overview/

随后使用Read工具查看这些jpg文件。

2. Identify Regions of Interest

2. 识别感兴趣区域

After viewing overview frames, identify timestamps where:

Something interesting is happening
More detail is needed
Action is occurring that requires temporal resolution

查看概览帧后，标记以下时间点：

有有趣内容发生
需要更多细节
有需要高时间分辨率的动作发生

3. Zoom In

3. 局部放大

For a time range (more frames, higher resolution):

bash

./skills/video-explorer/scripts/videx range <video> <start>-<end>

针对时间范围（更多帧，更高分辨率）：

bash

./skills/video-explorer/scripts/videx range <video> <start>-<end>

Example: videx range talk.mp4 5:30-6:00


**For higher temporal resolution** (catch fast action):
```bash
./skills/video-explorer/scripts/videx range <video> <start>-<end> --fps=10

For a single frame at full detail:

bash

./skills/video-explorer/scripts/videx zoom <video> <time>


**针对更高时间分辨率**（捕捉快速动作）：
```bash
./skills/video-explorer/scripts/videx range <video> <start>-<end> --fps=10

针对单帧全细节：

bash

./skills/video-explorer/scripts/videx zoom <video> <time>

Example: videx zoom talk.mp4 5:45

undefined

undefined

4. Iterate

4. 反复迭代

Repeat zoom operations as needed until the question is answered.

根据需要重复放大操作，直到问题得到解答。

Commands Reference

命令参考

Command	Purpose	Output
`videx overview <video>`	Quick timeline scan	320px frames @ 10s intervals
`videx overview <video> 5 480`	Denser timeline	480px frames @ 5s intervals
`videx range <video> <start>-<end>`	Extract segment	1280px frames @ 2fps
`videx range <video> <start>-<end> --fps=10`	Fast action	1280px frames @ 10fps
`videx zoom <video> <time>`	Single frame detail	1920px single frame
`videx zoom <video> <time> --hd`	Maximum detail	Full resolution frame

命令	用途	输出
`videx overview <video>`	快速时间线扫描	320像素帧，间隔10秒
`videx overview <video> 5 480`	更密集的时间线	480像素帧，间隔5秒
`videx range <video> <start>-<end>`	提取片段	1280像素帧，帧率2fps
`videx range <video> <start>-<end> --fps=10`	捕捉快速动作	1280像素帧，帧率10fps
`videx zoom <video> <time>`	单帧细节	1920像素单帧
`videx zoom <video> <time> --hd`	最大细节	全分辨率帧

Time Formats

时间格式

All commands accept these time formats:

```
1:30
```
= 1 minute 30 seconds
```
01:30:00
```
= 1 hour 30 minutes
```
90
```
= 90 seconds
```
1:30.5
```
= 1 minute 30.5 seconds

所有命令支持以下时间格式：

```
1:30
```
= 1分30秒
```
01:30:00
```
= 1小时30分
```
90
```
= 90秒
```
1:30.5
```
= 1分30.5秒

Output Structure

输出结构

Frames are saved with timestamps in filenames for easy reference:

./videx-out/
└── video-name/
    ├── overview/
    │   ├── t_00-00-00.00.jpg
    │   ├── t_00-00-10.00.jpg
    │   └── ...
    ├── range_5-30_6-00/
    │   ├── t_00-05-30.00.jpg
    │   └── ...
    └── zoom/
        └── t_00-05-45.00.jpg

帧以包含时间戳的文件名保存，便于参考：

./videx-out/
└── video-name/
    ├── overview/
    │   ├── t_00-00-00.00.jpg
    │   ├── t_00-00-10.00.jpg
    │   └── ...
    ├── range_5-30_6-00/
    │   ├── t_00-05-30.00.jpg
    │   └── ...
    └── zoom/
        └── t_00-05-45.00.jpg

Example Session

示例会话

User asks: "What happens in this lecture video around the 10 minute mark?"

Run overview to understand video structure
View overview frames, note that slides change around 9:30-11:00
Extract that range:
```
videx range lecture.mp4 9:30-11:00
```
View range frames, find the specific slide transition at 10:15
Zoom for detail:
```
videx zoom lecture.mp4 10:15
```
Report findings with specific timestamps

用户提问：“这个讲座视频10分钟左右发生了什么？”

运行概览命令以了解视频结构
查看概览帧，注意到9:30-11:00期间幻灯片发生切换
提取该时间段：
```
videx range lecture.mp4 9:30-11:00
```
查看该时间段的帧，找到10:15处的具体幻灯片切换
放大查看细节：
```
videx zoom lecture.mp4 10:15
```
结合具体时间点报告发现结果