README.md

---
title: README
emoji: 🔥
colorFrom: blue
colorTo: blue
sdk: static
pinned: false
---


# 🌟 Who We Are  

**DeepGHS** (*Deep Generative anime Hobbyist Syndicate*) is a passion-driven, non-profit community building anime/2D-focused infrastructure—because even otakus deserve robust tooling! 🎨  

We obsess over **ALL things anime-tech**:  
- 🖼️ Multimodal datasets (images, video, text, audio, 3D)  
- 🤖 AI models that actually *get* anime aesthetics  
- 🔧 Developer tools to make weeb R&D 10x easier  

**Our crew** includes:  
- **narugo1992** (self-proclaimed Jerry Mouse 🐭 | Software Engineering PhD | 25+ years of coding/teamwork wizardry)  
- A global squad of AI nerds + anime addicts from top labs/universities 🎓  
- Pro artists who know "moe" isn’t just a typo 🎨  

*Fun fact: We hire based on* ***passion for anime*** *first, technical skills second. (Sorry, resume bots!)*  

**Why trust us?**  
✅ 100% non-profit | ✅ Fully open-source | ✅ Zero paywalls (forever!)  
💖 Backed by sponsors & collaborators who *get* our vision  

---

# 🚀 Our Projects

## 📊 Datasets

### **Danbooru2024 Series**  
- **Original**: [Full-resolution images](https://huggingface.co/datasets/deepghs/danbooru2024)  
- **Compressed**: [WebP format (4M pixels)](https://huggingface.co/datasets/deepghs/danbooru2024-webp-4Mpixel)  
- **SFW Edition**: [Safe-for-work subset](https://huggingface.co/datasets/deepghs/danbooru2024-sfw)  

### **Sankaku Collection**  
- **Original**: [Full dataset](https://huggingface.co/datasets/deepghs/sankaku_full)  
- **Compressed**: [WebP format (4M pixels)](https://huggingface.co/datasets/deepghs/sankaku-webp-4Mpixel)  

### **Other Booru-style Repositories**  
- [Gelbooru Full](https://huggingface.co/datasets/deepghs/gelbooru_full)  
- [Yande.re Full](https://huggingface.co/datasets/deepghs/yande_full)  
- [Rule34.xxx Full](https://huggingface.co/datasets/deepghs/rule34_full)  
- [Konachan Full](https://huggingface.co/datasets/deepghs/konachan_full)  
- [Anime-Pictures Dataset](https://huggingface.co/datasets/deepghs/anime_pictures_full)  
- [Zerochan Full](https://huggingface.co/datasets/deepghs/zerochan_full)  

### **Functional Datasets**  
- [Character Similarity Dataset](https://huggingface.co/datasets/deepghs/character_similarity) - For cross-character recognition  
- [Anime Face Detection](https://huggingface.co/datasets/deepghs/anime_face_detection) - Labeled facial data  
- [Anime Head Detection](https://huggingface.co/datasets/deepghs/anime_head_detection) - Head position annotations  

### **Anime Series Datasets**  
- [BangumiBase](https://huggingface.co/BangumiBase) - Character-centric frame extractions  

### **Search Systems**  
- [Reverse Image Search](https://huggingface.co/spaces/deepghs/search_image_by_image) 🔍  
- [Danbooru Character Lookup](https://huggingface.co/spaces/deepghs/danbooru_character_search) 🕵️♂️  

---

## 🤖 Models

### **Classification Models**  
- [Anime/Real Classifier](https://huggingface.co/deepghs/anime_real_cls) - Distinguish anime from real images  
- [Image Type Classifier](https://huggingface.co/deepghs/anime_classification) - Categorize artwork styles  
- [Furry Detection](https://huggingface.co/deepghs/anime_furry) - Identify anthropomorphic characters  
- [Aesthetic Scorer](https://huggingface.co/deepghs/anime_aesthetic) - Predict visual appeal  
- [Style Era Classifier](https://huggingface.co/deepghs/anime_style_ages) - Decade recognition  
- [Live Demo](https://huggingface.co/spaces/deepghs/anime_image_classification) 🧪  

### **Detection Models**  
- [Face Detection](https://huggingface.co/deepghs/anime_face_detection) - Precise facial localization  
- [Head Detection](https://huggingface.co/deepghs/anime_head_detection) - Head position estimation  
- [Person Detection](https://huggingface.co/deepghs/anime_person_detection) - Full-body recognition  
- [NSFW Censor](https://huggingface.co/deepghs/anime_censor_detection) - Content moderation  
- [Detection Demo](https://huggingface.co/spaces/deepghs/anime_object_detection) 🎯  

### **Specialized Models**  
- [CCIP](https://huggingface.co/deepghs/ccip) - Character similarity encoding ([Demo](https://huggingface.co/spaces/deepghs/ccip))  
- [WD Tagger Enhanced](https://huggingface.co/deepghs/wd14_tagger_with_embeddings) - Tagger with embeddings  

---

## 🛠️ Open-Source Projects

### **Core Libraries**  
- [dghs-imgutils](https://github.com/deepghs/imgutils) - Foundational anime image processing toolkit  
- [waifuc](https://github.com/deepghs/waifuc) - Python framework for building data pipelines  
- [sdeval](https://github.com/deepghs/sdeval) - Quantitative evaluation for Stable Diffusion outputs  

### **Utilities**  
- [hfutils](https://github.com/deepghs/hfutils) - Enhanced HuggingFace Hub interface  
- [cheesechaser](https://github.com/deepghs/cheesechaser) - Targeted dataset sampling tool  

### **Training Systems**  
- [cyberharem](https://github.com/deepghs/cyberharem) - Automated LoRA training pipeline  

---

# 📬 Let’s Connect!  

**For collabs, sponsorships, or just anime-tech chatter**: 

- 💬Discord: [https://discord.gg/EAW4WqFdKY](https://discord.gg/EAW4WqFdKY)  (We don’t bite—unless you’re a bug!)  
- 🤗HuggingFace: [https://huggingface.co/deepghs](https://huggingface.co/deepghs)  
- 💻GitHub: [https://github.com/deepghs](https://github.com/deepghs)  
- 📧Email: [narugo1992@deepghs.org](mailto:narugo1992@deepghs.org)  

---

# 🎯 Want to Join?  

**We want YOU if**:  
1. You can code *and* name 10 JoJo stands 💪  
2. You’re ready to build—not just consume—anime tech  

**How to apply**:  
- Pitch us via [Discord](https://discord.gg/EAW4WqFdKY)/[Email](mailto:narugo1992@deepghs.org) with:  
  - Your anime-tech portfolio  
  - What you’ll bring to our dojo 🥋  
- OR: Directly request to join our [HuggingFace Org](https://huggingface.co/deepghs) , with the same things above

*Note: We do light vetting to protect our community—it’s faster than a Naruto run, promise!*  

---

# 🔮 What’s Next?  

More tools, datasets, and *~magic~* in development!  
**Pro tip**: Watch our repos ⭐ + Join Discord 👀 = Never miss an update!  

~~Btw, the above magic was polished by deepseek-R1. Gotta say, this LLM is a copywriting wizard.~~