---
title: ReWatch
canonical_url: "https://www.modelscope.cn/datasets/zcccccz/ReWatch"
md_url: "https://www.modelscope.cn/datasets/zcccccz/ReWatch.md"
repository: zcccccz/ReWatch
last_updated: 2025-09-29
license: "Apache License 2.0"
storage_size: "951 MB"
downloads: 312
stars: 1
---
# ReWatch
> ReWatch - zcccccz 在 ModelScope 开源的数据集。ReWatch Dataset Our ReWatch-R1 project page: https://rewatch-r1.github.io/ Video Source Our video sources come from other open-source data. You need to download the original video from the following links.
zcccccz/ReWatch 是 ModelScope 魔搭社区上的数据集，存储大小 951 MB，采用 Apache License 2.0 许可。
- **Repository**: zcccccz/ReWatch
- **License**: Apache License 2.0
- **Storage size**: 951 MB
- **Downloads**: 312
- **Stars**: 1
- **Last updated**: 2025-09-29
Source: https://www.modelscope.cn/datasets/zcccccz/ReWatch
---
# ReWatch Dataset
Our ReWatch-R1 project page: https://rewatch-r1.github.io/
## Video Source
Our video sources come from other open-source data. You need to download the original video from the following links.
VideoEspresso: https://huggingface.co/datasets/hshjerry0315/VideoEspresso_train_video \
Video-R1: https://huggingface.co/datasets/Video-R1/Video-R1-data \
MiraData: https://huggingface.co/datasets/TencentARC/MiraData \
Vript: https://huggingface.co/datasets/Mutonix/Vript \
VideoMarathon: https://huggingface.co/datasets/jylins/videomarathon
When using our data, you need to prepend the root directory of the video source to each video path:
```python
import os
import json
video_dirs = {
"VideoEspresso": "/path/to/your.../VideoEspresso_train_video",
"Video-R1": "/path/to/your.../Video-R1",
"MiraData": "/path/to/your.../mira_data",
"Vript": "/path/to/your.../vript_long_videos",
"VideoMarathon": "/path/to/your.../VideoMarathon"
}
caption_file = "caption.json"
caption_data = json.load(open(caption_file, "r"))
for item in caption_data:
source_name = item["source"] # video source
video_dir = video_dirs[source_name] # The root directory corresponding to the video source
ral_video_path = item["video_path"] # Relative path
abs_video_path = os.path.join(video_dir, video_dir) # Absolute path
```
For videos from MiraData, we use video_id and clip_id to determine the path of a video. You can find the video_id and clip_id of each video in the official repository of MiraData. For example, in our dataset, a video path from MiraData is:
```python
video_id = "ubgB1yelZMY"
clip_id = "49850.19"
video_path_in_our_dataset = f"{video_id}/{clip_id}.mp4"
```
For videos from VideoMarathon, we use their id in the VideoMarathon dataset to determine the path of a video. You need to use this id to find the download path of the original video in the VideoMarathon dataset.
For example, in our dataset, a video path from VideoMarathon is:
```python
id = "videomarathon_03293135"
video_path_in_our_dataset = f"{id}.mp4"
```
## Data Format
### Caption
```json
{
"video_path": "Youcook2/merged/4/6nVIgasiUtw.mkv",
"duration": 227,
"caption": "[00:00-00:05] The video opens with a ...",
"caption_summary": "[00:00-00:05] The video introduces ...",
"source": "VideoEspresso"
}
```
### QA
We use "question_id" to uniquely determine a question, The format are ```{video_path}_{question_type}_multiple_choice```or```{video_path}_{question_type}_open_end```
#### Multiple Choice
```json
{
"question_id": "4Z4E0IBaaRg.mp4_numerical_reasoning_multiple_choice",
"multiple_choice": true,
"question_type": "numerical_reasoning",
"question": "By how many seconds does the time interval between the player's second and third deaths exceed the time interval between the player's first and second deaths?\nA: 5 seconds\nB: 2 seconds\nC: 7 seconds\nD: 1 second",
"question_without_options": "By how many seconds does the time interval between the player's second and third deaths exceed the time interval between the player's first and second deaths?",
"mc_options": {
"A": "5 seconds",
"B": "2 seconds",
"C": "7 seconds",
"D": "1 second"
},
"answer": "D",
"video_path": "4Z4E0IBaaRg.mp4",
"source": "Vript"
}
```
#### Open End
```json
{
"question_id": "4Z4E0IBaaRg.mp4_numerical_reasoning_open_end",
"multiple_choice": false,
"question_type": "numerical_reasoning",
"question": "By how many seconds does the time interval between the player's second and third deaths exceed the time interval between the player's first and second deaths?",
"answer": "1 second",
"video_path": "4Z4E0IBaaRg.mp4",
"source": "Vript"
}
```
### CoT
react_chain is the interaction trajectory between the agent and the video, and natural_chain is the CoT data we use for training.
```json
{
"question_id": "XD-Violence/2/Kingsman.The.Golden.Circle.2017__#02-12-27_02-14-03_label_A.mp4_object_recognition_open_end",
"react_chain": [
{
"type": "thought",
"number": 1,
"content": "The question asks about named significant characters present among the guests at the grand church ceremony, aside from Eggsy, Princess Tilde, and Harry, and a distinct article of clothing worn by each. To answer this, I first need to locate the grand church ceremony in the video and identify the guests. I will start by retrieving the segment where the grand church ceremony occurs."
},
{
"type": "action",
"number": 1,
"content": "segment_retrieval(focus_event=\"the grand church ceremony, specifically the wedding ceremony of Eggsy and Princess Tilde\")"
},
{
"type": "observation",
"number": 1,
"content": "start_time: 00:53 \nend_time: 01:36"
},
{
"type": "thought",
"number": 2,
"content": "I have identified the time segment for the grand church ceremony. Now, I need to identify the significant characters present among the guests, excluding Eggsy, Princess Tilde, and Harry. I will query the segment focusing on the guests to identify them and their clothing. Since the question asks for named characters, I will focus on recognizable faces that might be identified as specific characters from the franchise."
},
{
"type": "action",
"number": 2,
"content": "segment_query(start_time=\"00:53\", end_time=\"01:36\", focus_aspect=\"guests and their distinct clothing, particularly looking for recognizable characters\")"
},
{
"type": "observation",
"number": 2,
"content": "Between 00:53 and 01:36, the wedding guests include notable characters such as Agent Tequila (Channing Tatum), dressed in a denim jacket, and Agent Ginger (Halle Berry), wearing a black dress. The congregation is diverse and features other significant figures from the film's universe in distinct, recognizable attire."
},
{
"type": "thought",
"number": 3,
"content": "The previous observation already identified two named significant characters, Agent Tequila and Agent Ginger, and described a distinct article of clothing for each. This information directly answers the question."
},
{
"type": "answer",
"content": "The two other named significant characters present among the guests are Agent Tequila, who is dressed in a denim jacket, and Agent Ginger, who is wearing a black dress.",
"number": 4
}
],
"natural_chain": "To answer this, I first need to locate the grand church ceremony in the video and identify the guests. I will start by retrieving the segment where the grand church ceremony occurs. I'll look for the segment focusing on the wedding ceremony of Eggsy and Princess Tilde. I've identified the ceremony between 00:53 and 01:36. Now, I need to identify the significant characters present among the guests, excluding Eggsy, Princess Tilde, and Harry. I will query the segment focusing on the guests to identify them and their clothing. Since the question asks for named characters, I will focus on recognizable faces that might be identified as specific characters from the franchise. I'll query the segment from 00:53 to 01:36, specifically looking for guests and their distinct clothing, particularly recognizable characters. I see Agent Tequila, dressed in a denim jacket, and Agent Ginger, wearing a black dress. This information directly answers the question. The two other named significant characters present among the guests are Agent Tequila, who is dressed in a denim jacket, and Agent Ginger, who is wearing a black dress."
}
```