README.md

metadata

language:
  - vi
license: cc-by-4.0
size_categories:
  - 1K<n<10K
task_categories:
  - video-classification
tags:
  - sign-language
  - vietnamese-sign-language
  - computer-vision
  - video-classification
configs:
  - config_name: default
    data_files:
      - split: train
        path: dataset_metadata.csv

AI Challenge CV Dataset Description

This dataset comes from https://aichallenge.ptit.edu.vn/ and is organized for a multi-class video classification task focusing on Vietnamese Sign Language.

---

🛠 Dataset Viewer Configuration

This dataset supports Hugging Face Dataset Viewer. The dataset structure and video paths are mapped via dataset_metadata.csv.

Expected CSV Structure

To ensure the Dataset Viewer renders correctly, your dataset_metadata.csv should follow this format:

video_file	label
train/Ăn/video_001.mp4	Ăn
train/Ban đêm/video_002.mp4	Ban đêm

---

Dataset Structure

The train folder contains training data organized in a class-based directory structure.

• each subfolder corresponds to a class label for the classification problem
• each subfolder name is the class label, for example: Ăn, Ban đêm, Bạn thân, Bệnh viện, Virus, Xe đạp, Xin lỗi, ...
• each folder contains many short MP4 video clips

What this means

• Each subfolder holds examples belonging to that label.
• The dataset is arranged as a label-directory dataset, which is ideal for multi-class classification tasks.
• Although the training set is organized by folder labels, the actual examples are short video files in MP4 format.

Representative Labels

Some notable class labels in train include:

• Ăn
• Ăn mừng
• An ủi
• Áp dụng
• Ban đêm
• Bạn thân
• Bệnh viện
• Bộ y tế
• Con gấu
• Cứu
• Học sinh
• Khẩu trang
• Lây bệnh
• Nghe
• Nghỉ ngơi
• Phòng xét nghiệm
• Sốt
• Sử dụng
• Thức ăn
• Thương
• Xe đạp
• Ô tô
• Xin lỗi
• Xuất viện
• ...

Summary

In summary, the training dataset is organized by label directories and is well suited for multi-class classification problems. For a deeper inspection, open each label subfolder to see the number of samples and the exact file formats inside.

Sample Visualizations

Sample Video Frames

Video Overview

Label Balance

Notes for Hugging Face

This description is intended for use on Hugging Face as an overview of the dataset structure and class organization. Ensure the dataset files and images are uploaded together so these images render correctly in the dataset card.