File size: 6,356 Bytes
e4ecf13 fb2e0da 1de9a10 2b502b1 cf090ae 1de9a10 2b502b1 dec16ec 1de9a10 cf090ae d51b11a cf090ae d51b11a cf090ae |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 |
---
license: apache-2.0
datasets:
- strangerguardhf/NSFW-MultiDomain-Classification-v2.0
language:
- en
base_model:
- google/siglip2-base-patch16-512
pipeline_tag: image-classification
library_name: transformers
tags:
- SigLIP2
- art
- explicit-content-detection
- media-filter
- Anime
---
# **siglip2-mini-explicit-content**
> **siglip2-mini-explicit-content** is an image classification vision-language encoder model fine-tuned from **`siglip2-base-patch16-512`** for a single-label classification task. It is designed to classify images into categories related to explicit, sensual, or safe-for-work content using the **SiglipForImageClassification** architecture.
> \[!Note]
> This model is intended to promote positive, safe, and respectful digital environments. Misuse is strongly discouraged and may violate platform or regional guidelines. As a classification model, it does not generate unsafe content and is suitable for moderation purposes.
> [!note]
*SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features* https://arxiv.org/pdf/2502.14786
> [!Important]
Note: Explicit, sensual, and pornographic content may appear in the results; however, all of them are considered not safe for work.
```py
Classification Report:
precision recall f1-score support
Anime Picture 0.8897 0.8296 0.8586 5600
Extincing & Sensual 0.8984 0.9477 0.9224 5618
Hentai 0.8993 0.9118 0.9055 5600
Pornography 0.9527 0.9285 0.9404 5970
Safe for Work 0.8957 0.9172 0.9063 6000
accuracy 0.9074 28788
macro avg 0.9072 0.9069 0.9066 28788
weighted avg 0.9076 0.9074 0.9071 28788
```
---
The model categorizes images into five classes:
* **Class 0:** Anime Picture
* **Class 1:** Extincing & Sensual
* **Class 2:** Hentai
* **Class 3:** Pornography
* **Class 4:** Safe for Work
---
# **Run with Transformers**
```python
!pip install -q transformers torch pillow gradio
```
```python
import gradio as gr
from transformers import AutoImageProcessor, SiglipForImageClassification
from transformers.image_utils import load_image
from PIL import Image
import torch
# Load model and processor
model_name = "prithivMLmods/siglip2-mini-explicit-content"
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)
# Updated labels
labels = {
"0": "Anime Picture",
"1": "Extincing & Sensual",
"2": "Hentai",
"3": "Pornography",
"4": "Safe for Work"
}
def detect_explicit_content(image):
"""Predicts content category in an uploaded image."""
image = Image.fromarray(image).convert("RGB")
inputs = processor(images=image, return_tensors="pt")
with torch.no_grad():
outputs = model(**inputs)
logits = outputs.logits
probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist()
predictions = {labels[str(i)]: round(probs[i], 3) for i in range(len(probs))}
return predictions
# Gradio Interface
iface = gr.Interface(
fn=detect_explicit_content,
inputs=gr.Image(type="numpy"),
outputs=gr.Label(label="Prediction Scores"),
title="siglip2-mini-explicit-content",
description="Upload an image to classify it as Anime, Hentai, Sensual, Pornographic, or Safe for Work."
)
if __name__ == "__main__":
iface.launch()
```
---
# **Guidelines for Use of siglip2-mini-explicit-content**
This model is designed for responsible content moderation and filtering. It is especially tuned for anime, hentai, and adult content. Use it ethically, with the following guidelines:
### **Recommended Use Cases**
* Content Moderation in social media and forums
* Parental Controls for safer browsing environments
* Dataset Curation for removing NSFW images from training data
* Safe Search Filtering for engines and discovery systems
* Workplace Image Scanning for compliance
### **Prohibited or Discouraged Use**
* Harassment, exposure, or targeting of individuals
* Use on private content without consent
* Illegal or unethical surveillance
* Sole reliance for legal or reputational decisions
* Deceptive manipulation of moderation results
---
# **Important Notes**
* Optimized for **anime and adult content detection**. Not suitable for detecting violence, drugs, or hate symbols.
* Probabilistic outputs — always **verify** with human review where needed.
* This model's predictions are **not legal classifications**.
---
## Demo Inference
> [!warning]
Anime Picture
> [!warning]
Extincing & Sensual
> [!warning]
Hentai
> [!warning]
Pornography
> [!warning]
Safe for Work
---
## **Ethical Reminder**
This tool was created to **enhance digital safety**. Do not use it to harm, surveil, or exploit individuals or communities. By using this model, you commit to ethical and privacy-respecting practices. |