🧠 Meme Filter AI

A lightweight AI-powered meme filtering tool that retrieves and analyzes memes using a combination of image and text features. It aims to support text-based meme retrieval (e.g., search by keywords like "crying girl") to find the most relevant memes from a dataset.

---

📌 Overview

This notebook:

• Downloads memes from the Imgflip API.
• Extracts and processes text (if present) from memes.
• Embeds meme content using CLIP (Contrastive Language–Image Pre-training) from OpenAI.
• Supports search via natural language queries (e.g., "spongebob meme" or "crying baby").
• Evaluates relevance by comparing CLIP text/image embeddings.
• Shows a comparison for search with both memes + text and search with only memes.

⚠️ Current limitation: The Imgflip API mostly provides memes without overlaid text, which reduces performance. To get the best results, use memes with visible text.

---

⚙️ Features

• ✅ Meme scraping with requests
• ✅ CLIP model usage for vision-language understanding
• ✅ Cosine similarity-based ranking
• ✅ Basic search functionality to find memes by text
• ✅ Colab-ready (includes folder creation, saving assets, etc.)

---

🚀 Usage

1. Download the Notebook

   • Clone this repo or download memes.ipynb.

2. Or Run in Google Colab

   • Upload the notebook to Google Colab.
   • All meme images will be saved to content/memes.

3. Run all cells

   • It will:
     • Fetch memes from the web
     • Extract features via CLIP
     • Allow you to search using keywords

4. Run Without Text Search

   • It will:
     • Fetch the memes Collected from API
     • Will Search Based on Image Alone
     • Allow you to search using keywords
     • Gives mediocre results

5. Run Text Search

   • It will:
     • Fetch the memes Collected from Drive
     • Will Search on Multi Modal Data (Image + Text)
     • Allow you to search using keywords
     • Gives Proper results

---

🔍 Sample Search Queries

Try searches like:

• "crying girl"
• "angry cat"
• "SpongeBob"
• "funny office"

The AI model really likes SpongeBob memes. 😄

---

🧪 Future Work

• Improve OCR to extract overlaid text from image.
• Support additional meme sources (e.g., Reddit).
• Deploy a minimal web interface.

---

📁 Folder Structure