msr-vtt

### Clone with HTTP
```bash
git clone https://www.modelscope.cn/datasets/AI-ModelScope/msr-vtt.git
```

### Load dataset
#### 1. Load video-meta data
```shell
from modelscope import MsDataset
ds = MsDataset.load('AI-ModelScope/msr-vtt', split='train')
print(next(iter(ds)))

Example:
sample = {'video_id': 'video0', 'caption': 'a car is shown', 'sen_id': 77300, 'category': 9, 'url': 'https://www.youtube.com/watch?v=9lZi22qLlEo', 'start time': 137.72, 'end time': 149.44, 'split': 'train', 'id': 0, '__index_level_0__': 0}

```

#### 2. Download videos
```shell
# Refer to: https://cove.thecvf.com/datasets/839

wget https://www.robots.ox.ac.uk/~maxbain/frozen-in-time/data/MSRVTT.zip

cd MSRVTT
unzip MSRVTT.zip
```

#### 3. Read local videos
```shell
import os

base_path = /path/to/MSRVTT/videos/all # Specify your own path
for sample in ds: # ds -- refer to above
video_id = sample['video_id']
video_path = os.path.join(base_path, f'{video_id}.mp4')

# process video
...

```

### 通过 HTTP 克隆
bash
git clone https://www.modelscope.cn/datasets/AI-ModelScope/msr-vtt.git

### 加载数据集
#### 1. 加载视频元数据
shell
from modelscope import MsDataset
ds = MsDataset.load('AI-ModelScope/msr-vtt', split='train')
print(next(iter(ds)))

示例：
样本格式如下：
{'video_id': 'video0', 'caption': '画面中展示了一辆汽车', 'sen_id': 77300, 'category': 9, 'url': 'https://www.youtube.com/watch?v=9lZi22qLlEo', 'start time': 137.72, 'end time': 149.44, 'split': '训练集', 'id': 0, '__index_level_0__': 0}

#### 2. 下载视频
shell
# 参考来源：https://cove.thecvf.com/datasets/839

wget https://www.robots.ox.ac.uk/~maxbain/frozen-in-time/data/MSRVTT.zip

cd MSRVTT
unzip MSRVTT.zip

#### 3. 读取本地视频
shell
import os

base_path = /path/to/MSRVTT/videos/all # 请指定你的本地路径
for sample in ds: # ds 指代上文加载的数据集
video_id = sample['video_id']
video_path = os.path.join(base_path, f'{video_id}.mp4')

# 视频处理流程
...