SlowQZK123/EmoArt-130k

EmoArt: A Large-Scale Emotion-Annotated Artistic Dataset

 

Overview

EmoArt is a comprehensive, large-scale emotion-annotated artistic dataset containing 132,664 high-resolution artworks spanning 56 painting styles across 7 thematic categories. This dataset bridges the gap between visual art and emotional computing, enabling groundbreaking research in emotion-aware AI systems.

Key Statistics

• —📊 132,664 artworks with rich emotional annotations
• —🎨 56 distinct painting styles from global art traditions
• —🌍 7 thematic categories covering diverse artistic movements
• —🧠 Multi-dimensional emotion labeling with therapeutic insights
• —🤖 GPT-4o enhanced annotations with human validation

Dataset Versions

We provide two carefully curated versions to meet different research needs:

Quick Start

Download by Category

Access individual artistic categories as needed:

# Download specific category (e.g., Cubism)
wget https://huggingface.co/datasets/printblue/EmoArt-130k/resolve/main/Cubism.tar.gz
tar -xzvf Cubism.tar.gz

# Download all annotations (single file)
wget https://huggingface.co/datasets/printblue/EmoArt-130k/resolve/main/Annotation.json

Load with Python

from datasets import load_dataset

# Authenticate with Hugging Face
# Run: huggingface-cli login

# Load the complete dataset
ds = load_dataset("printblue/EmoArt-130k")

Dataset Architecture

File Organization

The dataset follows a modular structure for efficient access and storage:

EmoArt Dataset/
├── Classics.tar.gz              # Traditional art styles
├── Modern_Edge.tar.gz           # Modern movements  
├── East_Spirit.tar.gz           # Asian art traditions
├── Chromatic_Soul.tar.gz        # Color-focused styles
├── Dream_Visions.tar.gz         # Surreal and abstract works
├── Form_Flow.tar.gz             # Minimalist and geometric styles
├── Social_Mirror.tar.gz         # Social and political art
└── Annotation.json              # Unified annotation file

Architecture Benefits:

• —🔄 Modular downloads: Access only needed categories
• —📁 Centralized annotations: Single JSON file for all metadata
• —💾 Storage efficient: Compressed tar.gz format
• —🚀 Fast experimentation: Category-specific access

Individual Category Structure

Each category archive contains organized image files:

{Category}.tar.gz
└── images/
    ├── image_00001.jpg
    ├── image_00002.jpg
    ├── image_00003.jpg
    └── ... (category-specific count)

Annotation Framework

Comprehensive Annotation Schema

The unified Annotation.json provides rich metadata for every artwork:

{
    "request_id": "{Art_Category}_request-1",
    "description": {
        "first_section": {
            "description": "Detailed visual and compositional analysis of the artwork"
        },
        "second_section": {
            "visual_attributes": {
                "brushstroke": "Technical analysis of brushwork and application",
                "color": "Color palette, saturation, and harmony assessment",
                "composition": "Structural organization and visual flow analysis",
                "light_and_shadow": "Illumination, contrast, and depth evaluation",
                "line_quality": "Line characteristics and stylistic execution"
            },
            "emotional_impact": "Comprehensive emotional response and psychological effect analysis"
        },
        "third_section": {
            "emotional_arousal_level": "High/Low",
            "emotional_valence": "Positive/Negative", 
            "dominant_emotion": "Primary emotional category",
            "healing_effects": ["Therapeutic and wellness applications"]
        }
    },
    "image_path": "Images\\{Category}\\{filename}.jpg"
}

Annotation Dimensions

Each artwork includes multi-layered emotional and visual analysis:

Emotional Intelligence

• —12 Core Emotions: Joy, Sadness, Anger, Fear, Surprise, Disgust, Calm, Excitement, and more
• —Valence Classification: Positive/Negative emotional tone
• —Arousal Assessment: High/Low emotional intensity
• —Dominant Response: Primary emotional impact

Visual Analysis Framework

• —Brushstroke Technique: Application method and artistic execution
• —Color Psychology: Palette choices and emotional resonance
• —Compositional Structure: Visual organization and flow
• —Light Dynamics: Illumination patterns and contrast usage
• —Linear Expression: Line quality and stylistic character

Therapeutic Applications

• —Healing Potential: Stress relief, mood enhancement, contemplative benefits
• —Wellness Integration: Applications in art therapy and mental health

Artistic Categories

Thematic Organization

The dataset encompasses seven major artistic domains:

Style Coverage

The dataset represents the full spectrum of global artistic expression across cultures and time periods, ensuring comprehensive coverage for cross-cultural emotion research.

Performance Benchmarks

Emotion-Aware Generation Models

Evaluation results on state-of-the-art diffusion models:

Higher scores indicate better performance for Quality and Emotion Alignment; lower FID scores indicate better image quality.

Research Applications

Core Research Areas

• —Emotion-Aware AI: Training models that understand and generate emotionally resonant content
• —Affective Computing: Bridging human emotion and computational understanding
• —Cross-Modal Learning: Vision-language models with emotional intelligence
• —Computational Aesthetics: Quantifying beauty and emotional impact in art
• —Digital Art Therapy: AI-assisted therapeutic applications

Practical Applications

• —Content Generation: Emotion-driven artistic creation
• —Therapeutic Tools: AI-powered art therapy systems
• —Cultural Studies: Cross-cultural emotion perception research
• —Educational Technology: Interactive art history and emotion learning
• —Creative Industries: Emotion-aware design and marketing tools

Quality Assurance

Ethical Standards

• —✅ Open Access Only: Exclusively public domain and Creative Commons works
• —🔍 Content Filtering: Manual review for sensitive or inappropriate material
• —🌍 Cultural Balance: Representative sampling across global art traditions
• —👥 Human Oversight: Expert validation of AI-generated annotations

Technical Quality

• —🖼️ High Resolution: Professional-quality image standards
• —🤖 AI-Human Hybrid: GPT-4o annotations with human expert validation
• —📊 Consistency Checks: Standardized annotation protocols
• —🔄 Continuous Improvement: Community feedback integration

Getting Started

For Researchers

1. 1.Browse the dataset on Hugging Face to understand scope and structure
2. 2.Start with EmoArt-5k for initial experiments and prototyping
3. 3.Download specific categories relevant to your research focus
4. 4.Scale to EmoArt-130k for comprehensive model training

For Developers

1. 1.Use the Python API for seamless integration
2. 2.Implement modular loading to manage memory efficiently
3. 3.Leverage the annotation structure for multi-task learning
4. 4.Contribute improvements back to the community

Resources & Support

Documentation & Code

• —📖 Research Paper: MM'25 Conference Publication
• —💻 Source Code: GitHub Repository
• —📊 Benchmarks: Performance baselines and evaluation metrics

Community & Support

• —📧 Contact: zhangcheng2122@jlu.edu.com
• —🐛 Bug Reports: GitHub Issues
• —💬 Discussions: Hugging Face Community

"Art enables us to find ourselves and lose ourselves at the same time." EmoArt enables AI to do the same.