**Direct Download Link:**
https://osf.io/download/wdrht/
**Dataset Overview:**
LRW-AR consists of 100 word classes, each representing a word in Arabic. The dataset is structured into three subsets: training, validation, and testing. Each class contains 200 videos, distributed as follows:
- **Training Set:** 160 videos per class
- **Validation Set:** 20 videos per class
- **Test Set:** 20 videos per class
**Dataset Structure:**
The dataset follows organized by subsets and classes.
LRW-AR dataset is devided into 3 directories (train, val and test) which contains subdirectories for each class. Each class folder itself contain video files (videoN.mp4) and their corresponding annotation files (videoN.csv).
LRW-AR Dataset
│
├── Train
│ ├── Class 1
│ │ ├── video1.mp4
│ │ ├── video1.csv
│ │ ├── ...
│ │
│ ├── Class 2
│ │ ├── video1.mp4
│ │ ├── video1.csv
│ │ ├── ...
| .
.
|
│ └── Class 100
│ ├── video1.mp4
│ ├── video1.csv
│ ├── ...
│
├── Val
│ ├── ...
│
└── Test
├── ...
