app.py

# app.py

import numpy as np
import gradio as gr
from sklearn.cluster import KMeans
from transformers import (
    BlipProcessor,
    BlipForConditionalGeneration,
    MBartForConditionalGeneration,
    MBart50TokenizerFast,
)
from PIL import Image, ImageDraw
import requests
from io import BytesIO

# Download example images
def download_example_images():
    image_urls = [
        # URL format: ("Image Description", "Image URL")
        ("Sunset over Mountains", "https://images.unsplash.com/photo-1501785888041-af3ef285b470"),
        ("Forest Path", "https://images.unsplash.com/photo-1502082553048-f009c37129b9"),
        ("City Skyline", "https://images.unsplash.com/photo-1498598453737-8913e843c47b"),
        ("Beach and Ocean", "https://images.unsplash.com/photo-1507525428034-b723cf961d3e"),
        ("Desert Dunes", "https://images.unsplash.com/photo-1501594907352-04cda38ebc29"),
    ]
    
    example_images = []
    for idx, (description, url) in enumerate(image_urls, start=1):
        response = requests.get(url)
        if response.status_code == 200:
            img = Image.open(BytesIO(response.content))
            img.save(f'example{idx}.jpg')
            example_images.append([f'example{idx}.jpg'])
        else:
            print(f"Failed to download image from {url}")
    return example_images

# Download example images and prepare examples list
examples = download_example_images()

# Load and Process the Entire Image
def load_image(image):
    # Convert PIL image to numpy array (RGB)
    image_np = np.array(image.convert('RGB'))
    
    # Resize the image for better processing
    resized_image = image.resize((300, 300), resample=Image.LANCZOS)
    resized_image_np = np.array(resized_image)
    
    return resized_image_np

# Extract Dominant Colors from the Image
def extract_colors(image, k=8):
    # Flatten the image
    pixels = image.reshape(-1, 3)
    # Normalize pixel values to [0, 1]
    pixels = pixels / 255.0
    # Ensure data type is float64
    pixels = pixels.astype(np.float64)
    # Apply K-means clustering to find dominant colors
    kmeans = KMeans(n_clusters=k, random_state=0, n_init=10, max_iter=300)
    kmeans.fit(pixels)
    # Convert normalized colors back to 0-255 scale
    colors = (kmeans.cluster_centers_ * 255).astype(int)
    return colors

# Create an Image for the Color Palette
def create_palette_image(colors):
    num_colors = len(colors)
    palette_height = 100
    palette_width = 100 * num_colors
    palette_image = Image.new("RGB", (palette_width, palette_height))
    
    draw = ImageDraw.Draw(palette_image)
    for i, color in enumerate(colors):
        # Ensure color values are within the valid range and integers
        color = tuple(np.clip(color, 0, 255).astype(int))
        draw.rectangle([i * 100, 0, (i + 1) * 100, palette_height], fill=color)
    
    return palette_image

# Display Color Palette as Hex Codes
def display_palette(colors):
    hex_colors = []
    for color in colors:
        # Ensure color values are within valid range and integers
        color = np.clip(color, 0, 255).astype(int)
        hex_color = "#{:02x}{:02x}{:02x}".format(color[0], color[1], color[2])
        hex_colors.append(hex_color)
    return hex_colors

# Generate Image Caption Using Hugging Face BLIP
def generate_caption(image):
    # Load models only once
    if 'processor' not in generate_caption.__dict__:
        generate_caption.processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
        generate_caption.model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")
    processor = generate_caption.processor
    model = generate_caption.model
    
    inputs = processor(images=image, return_tensors="pt")
    output = model.generate(**inputs)
    caption = processor.decode(output[0], skip_special_tokens=True)
    return caption

# Translate Caption to Arabic Using mBART
def translate_to_arabic(text):
    # Load models only once
    if 'tokenizer' not in translate_to_arabic.__dict__:
        translate_to_arabic.tokenizer = MBart50TokenizerFast.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
        translate_to_arabic.model = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
    tokenizer = translate_to_arabic.tokenizer
    model = translate_to_arabic.model
    
    tokenizer.src_lang = "en_XX"
    encoded = tokenizer(text, return_tensors="pt")
    generated_tokens = model.generate(
        **encoded, 
        forced_bos_token_id=tokenizer.lang_code_to_id["ar_AR"]
    )
    translated_text = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
    return translated_text

# Gradio Interface Function (Combining Elements)
def process_image(image):
    # Ensure input is a PIL Image
    if isinstance(image, np.ndarray):
        image = Image.fromarray(image)
    
    # Convert to RGB format for PIL processing
    image_rgb = image.convert("RGB")
    
    # Load and resize the entire image
    resized_image_np = load_image(image_rgb)
    
    # Convert resized image to PIL Image for Gradio output
    resized_image_pil = Image.fromarray(resized_image_np)
    
    # Generate caption using BLIP model
    caption = generate_caption(image_rgb)
    
    # Translate caption to Arabic
    caption_arabic = translate_to_arabic(caption)
    
    # Extract dominant colors from the entire image
    colors = extract_colors(resized_image_np, k=8)
    color_palette = display_palette(colors)
    
    # Create palette image
    palette_image = create_palette_image(colors)
    
    # Combine English and Arabic captions
    bilingual_caption = f"English: {caption}\nArabic: {caption_arabic}"
    
    return bilingual_caption, ", ".join(color_palette), palette_image, resized_image_pil

# Create Gradio Interface using Blocks and add a submit button
with gr.Blocks(css=".gradio-container { height: 1000px !important; }") as demo:
    gr.Markdown("<h1 style='text-align: center;'>Palette Generator from Image with Image Captioning</h1>")
    gr.Markdown(
        """
        <p style='text-align: center;'>
        Upload an image or select one of the example images below to generate a color palette and a description of the image in both English and Arabic.
        </p>
        """
    )
    with gr.Row():
        with gr.Column(scale=1):
            image_input = gr.Image(type="pil", label="Upload your image or select an example below")
            submit_button = gr.Button("Submit")
            gr.Examples(
                examples=examples,
                inputs=image_input,
                label="Example Images",
                examples_per_page=5,
            )
        with gr.Column(scale=1):
            caption_output = gr.Textbox(label="Bilingual Caption", lines=5, max_lines=10)
            palette_hex_output = gr.Textbox(label="Color Palette Hex Codes", lines=2)
            palette_image_output = gr.Image(type="pil", label="Color Palette")
            resized_image_output = gr.Image(type="pil", label="Resized Image")
    
    submit_button.click(
        fn=process_image,
        inputs=image_input,
        outputs=[caption_output, palette_hex_output, palette_image_output, resized_image_output],
    )

# Launch Gradio Interface
demo.launch()