Updated the code

.gitignore

@@ -0,0 +1,8 @@
+.env
+./env
+./venv
+.venv
+.venv
+./venv
+./__pycache__
+.__pycache__

README.md

@@ -1,3 +1,53 @@
+# 🖼️ Image Sticker Generator API
+Welcome to the Image Sticker Generator API! This service is designed to transform images into "stickers" by isolating the main object with advanced semantic segmentation. The stickers are crafted with smooth, feathered edges to ensure a clean and professional appearance.
+
+
+## Examples 
+
+![Example1](output/sticker_check3.png)
+![Example2](output/sticker_check.png)
+![Example3](output/sticker_check5.png)
+
+
+## 📋 Features
+### 🔍 Semantic Segmentation: Automatically detects and highlights the main object in an image, ensuring accurate cutouts and clear subject isolation.
+### 🌟 Feathered Edges: Applies a soft Gaussian blur to mask edges, creating a natural transition to transparency.
+### ⚡ Built with FastAPI: Ensures high performance, scalability, and rapid response times for production use.
+### 📂 Versatile Image Support: Accepts both PNG and JPEG image formats.
+
+### 🚀 Quick Start Guide
+Clone the repository and navigate to the project directory.
+Install Dependencies with Python 3.8+.
+Start the API Server using the provided configuration.
+Once running, the API will be accessible locally, ready for image uploads and sticker creation.
+
+## 🛠️ API Endpoints
+POST /create_sticker/
+Description: Upload an image to generate a sticker with a transparent background.
+Supported File Types: Accepts PNG and JPEG formats.
+Response: Returns a PNG image of the sticker with a transparent background. In case of an unsupported file format, an error message will be returned.
+## 🧩 How It Works
+Model & Preprocessing: Uploaded images are preprocessed and passed through a pre-trained model for semantic segmentation.
+Mask Generation: A binary mask isolates the main object in the image.
+Edge Feathering: A Gaussian blur is applied to the mask edges to create a soft transition.
+Sticker Creation: The mask is used to add transparency, producing an image that can be directly used as a sticker.
+## 📂 Directory Structure
+The main application file manages the API and endpoints, while generated stickers are saved in a designated output directory. This ensures easy access and organization for generated images.
+
+## ⚙️ Configuration
+Before running the application, confirm that the output directory exists. This is essential for storing all generated stickers for easy retrieval and management.
+
+## 📜 License
+Licensed under the MIT License, making it easy for anyone to adapt and build upon the work.
+
+## 🙋‍♂️ Contributing
+Contributions are welcome! To contribute, create a new issue or pull request for bug fixes, enhancements, or new features. All contributions should adhere to the project's coding standards and guidelines.
+
+## Built with ❤️ by [SAMIULLAH]
+
+## 📞 Support
+For support or inquiries, please contact nicesami156@gmail.com.
+
 ---
 license: mit
 ---

main.py

@@ -0,0 +1,66 @@
+from fastapi.responses import FileResponse
+from transformers import AutoImageProcessor, AutoModelForSemanticSegmentation
+from PIL import Image, ImageFilter
+from fastapi import FastAPI, File, UploadFile, HTTPException
+import numpy as np
+import uvicorn
+import torch
+import os
+
+app = FastAPI()
+
+# Load the model and processor once to avoid reloading on every request
+preprocessor = AutoImageProcessor.from_pretrained("google/deeplabv3_mobilenet_v2_1.0_513")
+model = AutoModelForSemanticSegmentation.from_pretrained("google/deeplabv3_mobilenet_v2_1.0_513")
+
+# Create the directory for saving output if it doesn’t exist
+os.makedirs("output", exist_ok=True)
+
+def get_segmentation_mask(image: Image.Image) -> Image.Image:
+    """Generate a binary segmentation mask with feathered edges from an input image."""
+    # Step 1: Preprocess and run model inference
+    inputs = preprocessor(images=image, return_tensors="pt")
+    with torch.no_grad():
+        outputs = model(**inputs)
+
+    # Step 2: Post-process the segmentation output to get a clean binary mask
+    predicted_mask = preprocessor.post_process_semantic_segmentation(outputs)[0]
+    mask_np = predicted_mask.cpu().numpy().astype("uint8") * 255  # Convert to binary values (0 or 255)
+    binary_mask = Image.fromarray(mask_np)
+
+    # Step 3: Apply a slight Gaussian blur to soften the edges
+    feathered_mask = binary_mask.filter(ImageFilter.GaussianBlur(1))  # Adjust blur radius as needed
+    feathered_mask = feathered_mask.resize(image.size, Image.BICUBIC)
+
+    return feathered_mask
+
+def apply_mask_to_image(image: Image.Image, mask: Image.Image) -> Image.Image:
+    """Apply the segmentation mask to the input image to create a transparent sticker."""
+    image = image.convert("RGBA")  # Ensure image is in RGBA mode
+    sticker = Image.new("RGBA", image.size)
+    sticker.paste(image, (0, 0), mask)  # Use mask as the alpha channel
+    return sticker
+
+@app.post("/create_sticker/")
+async def create_sticker(file: UploadFile = File(...)):
+    """Endpoint to convert an uploaded image to a sticker."""
+    if file.content_type not in ["image/png", "image/jpeg"]:
+        raise HTTPException(status_code=400, detail="Invalid image format. Only PNG and JPEG are supported.")
+
+    # Load the image
+    input_image = Image.open(file.file).convert("RGB")
+
+    # Generate segmentation mask and apply it to create a sticker
+    mask = get_segmentation_mask(input_image)
+    sticker = apply_mask_to_image(input_image, mask)
+
+    # Save the output sticker
+    output_path = f"output/sticker_{file.filename}"
+    sticker.save(output_path, "PNG")
+
+    return FileResponse(output_path, media_type="image/png")
+
+
+# Run the app
+if __name__ == "__main__":
+    uvicorn.run(app, host="0.0.0.0", port=8000)

requirements.txt

@@ -0,0 +1,15 @@
+diffusers 
+transformers 
+accelerate
+huggingface_hub
+pillow
+fastapi
+pydantic 
+ngrok 
+uvicorn
+celery 
+redis
+sentencepiece
+tokenizers
+opencv-python
+python-multipart