license: cc-by-nc-4.0
task_categories:
- video-classification
tags:
- aesthetics
- video-quality
- multimedia
language:
- en
This repository introduces VADB, a large-scale video aesthetics database, presented in the paper VADB: A Large-Scale Video Aesthetic Database with Professional and Multi-Dimensional Annotations.
The VADB dataset is designed to provide comprehensive resources for researchers and developers interested in video aesthetics analysis, computer vision, and multimedia content assessment.
The associated code and models can be found on GitHub: https://github.com/BestiVictory/VADB
Dataset Details
The data from our video dataset is stored in VADB.zip.
The VADB dataset includes:
- 7,881 videos covering diverse video styles and content categories.
- Detailed language comments for each video.
- Aesthetic scores across 7-11 dimensions, comprehensively covering the aesthetic attribute features of videos.
- Rich objective tags, annotating video shooting techniques and other objective dimensions.
Our train.csv contains all annotated data related to aesthetic scores. For text annotations such as comments and tags mentioned in the paper, please refer to merged_comment_tag.json.
[2025/10/22 UPDATE] The annotation files are fully open-sourced, but due to copyright issues with some videos, 2,609 video clips have been retained and not open-sourced, while the remaining 7,881 video clips have been fully released.
Getting Started & Sample Usage
🚀 Install Dependencies
First, install the required dependencies (From CLIP):
conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0
pip install ftfy regex tqdm
pip install opencv-python boto3 requests pandas
📦 Download the Dataset
Download the VADB dataset from Hugging Face:
git lfs install
git clone https://huggingface.co/datasets/BestiVictoryLab/VADB
🧠 VADB-Net Scoring Models
The VADB project also provides VADB-Net, a novel video aesthetics scoring framework. You will need to load a pre-trained video encoder model (details and link provided in the GitHub repository's Video Encoder section). This encoder extracts aesthetic feature vectors from videos and serves as the foundational component for all scoring models.
The repository is structured into three main components for different scoring tasks:
1. Overall Aesthetic Score
- Folder:
1TotalScore - Model: Predicts the overall aesthetic score of videos
- Usage:
cd 1TotalScore python 1TotalScore.py
2. General Attribute Scores
- Folder:
2GeneralAttribute - Model: Evaluates general aesthetic attributes of videos
- Evaluation Dimensions: Composition, Shot Size, Lighting, Visual Tone, Color, Depth of Field
- Usage:
cd 2GeneralAttribute python 2GeneralAttribute.py
3. Human-Centric Attribute Scores
- Folder:
3HumanAttribute - Model: Focuses on evaluating specific aesthetic attributes of human subjects
- Evaluation Dimensions: Expression, Movement, Costume, Makeup
- Usage:
cd 3HumanAttribute python 3HumanAttribute.py