| --- |
| title: STAR |
| emoji: 😻 |
| colorFrom: red |
| colorTo: indigo |
| sdk: streamlit |
| sdk_version: 1.43.2 |
| app_file: app.py |
| pinned: false |
| license: mit |
| short_description: Secure Transformer-based Adult-content Recognition |
| --- |
| |
| # ⭐ STAR - Secure Transformer-based Adult-content Recognition |
|
|
| **STAR (Secure Transformer-based Adult-content Recognition)** is an advanced AI model designed for multi-label classification of adult content using deep learning techniques. Built on a **Swin Transformer** backbone, STAR aims to provide secure and explainable detection of explicit content while ensuring ethical AI deployment. |
|
|
| ## 🚀 Features |
|
|
| - **Multi-label Classification**: Detects and categorizes adult content into multiple subcategories. |
| - **Explainable AI (XAI)**: Integrates **Class Activation Mapping (CAM)** for visual interpretability. |
| - **Secure & Ethical**: Designed with responsible AI principles to prevent misuse. |
| - **Transformer-based Architecture**: Utilizes **Swin Transformer** for high accuracy in image classification. |
|
|
| --- |
|
|
| ## 📌 Content Classification |
|
|
| The model categorizes images into the following **main categories**, each with its respective **subcategories (tags):** |
|
|
| | **Main Category** | **Subcategories (Tags)** | |
| |---------------------|-------------------------| |
| | **Adult** | `in_bikini`, `bra_exposed`, `panty_exposed`, `breast_exposed`, `vagina_exposed`, `butt_exposed`, `penis_exposed` | |
| | **Real Nude** | `celebrity_nude`, `celebrity_movie`, `boudior`, `cp_nude` | |
| | **Fake Nude** | `celebrity_fake`, `cp_fake`, `revenge_porn`, `loan_app_defamation` | |
| | **Pornographic** | `masturbating`, `kissing`, `intercourse`, `oral`, `professional_porn`, `leaked_videos`, `amateur_porn` | |
| | **Artistic** | `painting`, `statue`, `digital_art`, `ai_gen` | |
|
|
| --- |
|
|
| ## 🛠️ Model Architecture |
|
|
| STAR is powered by **Swin Transformer**, a hierarchical vision transformer model known for its efficiency and scalability in image classification. The model processes images with **pretrained weights**, fine-tuned for explicit content recognition. |
|
|
| --- |
|
|
| ## 🏆 Ethical AI Commitment |
|
|
| 🔹 **STAR** is designed for **ethical AI research and secure content moderation**. |
| 🔹 The model is **not intended for censorship or privacy violations**. |
| 🔹 Users must **ensure compliance with local regulations** when deploying this model. |
|
|
| --- |
|
|
| ## 📜 License |
|
|
| This project is open-source and provided under the **MIT License**. Please use responsibly. |
|
|
| --- |
|
|
| ## ✉️ Contact |
|
|
| For inquiries or collaborations, reach out to **Ramaguru Radhakrishnan**. |
|
|
|
|
|
|
| Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference |
|
|
