🚀 FocalNet NSFW Image Classifier: Your Content Moderation Superhero! 🦸‍♂️

🌟 Discover the Power of Intelligent Moderation!

👋 Are you ready for a revolution in content moderation? Meet the FocalNet NSFW Image Classifier - your new, lightning-fast, and super-smart assistant in the battle against inappropriate content!

🎭 Who Am I?

I'm an advanced AI model, built on the powerful microsoft/focalnet-base. My superpower is the lightning-fast classification of images into three categories:

🟢 SAFE: "Green light! Let's roll with this content!"
🟡 QUESTIONABLE: "Hmm... Maybe we should take a second look?"
🔴 UNSAFE: "Whoa! Let's stop this before anyone sees it!"

🦾 What Can I Do?

Imagine you're the guardian of the internet galaxy. Your mission? Protect users from shocking, inappropriate content. But how do you review millions of images daily? That's where I come in!

🕵️‍♂️ Lightning-Fast Detection: I'll analyze every pixel faster than you can say "safe content"!
🛡️ Protective Shield: I'll stand guard over your platforms, shielding users from unwanted content.
🎯 Sniper's Precision: My eye is so sharp that I can spot potential threats with surgical accuracy.

🚀 How to Use Me?

Ready for an adventure? Here's how you can harness my power:

Install my powers:

pip install transformers==4.37.2 torch==2.3.1 torchvision Pillow

Summon me in your code:

import os
from PIL import Image
import torch
from torchvision import transforms
from transformers import AutoProcessor, FocalNetForImageClassification

# Path to the folder with images
image_folder = ""
# Path to the model
model_path = "MichalMlodawski/nsfw-image-detection-large" 

# List of jpg files in the folder
jpg_files = [file for file in os.listdir(image_folder) if file.lower().endswith(".jpg")]

# Check if there are jpg files in the folder
if not jpg_files:
    print("🚫 No jpg files found in folder:", image_folder)
    exit()

# Load the model and feature extractor
feature_extractor = AutoProcessor.from_pretrained(model_path)
model = FocalNetForImageClassification.from_pretrained(model_path)
model.eval()

# Image transformations
transform = transforms.Compose([
    transforms.Resize((512, 512)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

# Mapping from model labels to NSFW categories
label_to_category = {
    "LABEL_0": "Safe",
    "LABEL_1": "Questionable",
    "LABEL_2": "Unsafe"
}

# Processing and prediction for each image
results = []
for jpg_file in jpg_files:
    selected_image = os.path.join(image_folder, jpg_file)
    image = Image.open(selected_image).convert("RGB")
    image_tensor = transform(image).unsqueeze(0)
    
    # Process image using feature_extractor
    inputs = feature_extractor(images=image, return_tensors="pt")
    
    # Prediction using the model
    with torch.no_grad():
        outputs = model(**inputs)
        probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1)
        confidence, predicted = torch.max(probabilities, 1)
    
    # Get the label from the model's configuration
    label = model.config.id2label[predicted.item()]
    
    results.append((jpg_file, label, confidence.item() * 100))

# Display results
print("🖼️  NSFW Classification Results 🖼️")
print("=" * 40)
for jpg_file, label, confidence in results:
    category = label_to_category.get(label, "Unknown")
    emoji = {"Safe": "✅", "Questionable": "⚠️", "Unsafe": "🔞"}.get(category, "❓")
    confidence_bar = "🟩" * int(confidence // 10) + "⬜" * (10 - int(confidence // 10))
    
    print(f"📄 File name: {jpg_file}")
    print(f"🏷️ Model Label: {label}")
    print(f"{emoji} NSFW Category: {category}")
    print(f"🎯 Confidence: {confidence:.2f}% {confidence_bar}")
    print(f"{'=' * 40}")

print("🏁 Classification completed! 🎉")

🎉 What Sets Me Apart?

🚄 Speed of Light: I'll analyze thousands of images before you finish your morning coffee!
🧠 Intelligence Level 100: I've learned from millions of examples, so I know all the tricks!
🛠️ Easy Integration: I'll hop into your code faster than a cat on a keyboard!
🌐 Multilingual Support: I understand images from all cultures and contexts!
🔄 Continuous Learning: I'm always improving, adapting to new trends and challenges!

🔬 Technical Specifications

Base Model: microsoft/focalnet-base
Model Type: FocalNetForImageClassification
Input Size: 512x512 pixels
Output: 3 classes (Safe, Questionable, Unsafe)
Framework: PyTorch
Language: Python 3.6+

🚀 Use Cases

Social Media Platforms: Keep user-generated content clean and safe.
E-commerce Sites: Ensure product images meet community standards.
Dating Apps: Maintain a respectful environment for all users.
Content Sharing Platforms: Automatically filter potentially inappropriate uploads.
Educational Platforms: Ensure learning materials are age-appropriate.

🏋️ Training and Performance

Training Data: Millions of diverse images across various categories
Fine-tuning: Specialized NSFW dataset for precise categorization
Accuracy: 95%+ on benchmark NSFW detection tasks
Latency: <100ms per image on standard GPU hardware

⚠️ Important Warnings (Because Every Superhero Has Their Weaknesses)

🎢 Not for Extreme Challenges: I'm great, but don't use me where an error could cost more than burnt toast!
🤖 I'm Not Skynet: I can make mistakes sometimes, so don't leave me alone with the red button!
🕵️‍♂️ Respect Privacy: Make sure you have the right to process the images you show me. I don't like prying eyes!
🔄 I Need Updates: The world changes, and so must I! Regularly check if I need a refresh.
🤝 Collaboration is Key: I'm a great assistant, but let's leave final decisions to humans. Together, we're unbeatable!

🌈 The Future is Bright!

Remember, I'm part of an ongoing research process. With each update, I become smarter, faster, and even more incredible!

Ready to revolutionize content moderation together? Bring me on board your project and watch the magic happen! 🎩✨

Join the AI revolution today and make the internet a safer place! 🌍💪

📚 References and Resources

Let's make the digital world safer, one image at a time! 🌟

MichalMlodawski
/

nsfw-image-detection-large