import os from typing import Tuple import gradio as gr import numpy as np import pandas as pd from PIL import Image from sklearn.cluster import KMeans def _image_resize(image: Image.Image, pixels: int = 90000, **kwargs): rt = (image.size[0] * image.size[1] / pixels) ** 0.5 if rt > 1.0: small_image = image.resize((int(image.size[0] / rt), int(image.size[1] / rt)), **kwargs) else: small_image = image.copy() return small_image def get_main_colors(image: Image.Image, n: int = 28, pixels: int = 90000) \ -> Tuple[Image.Image, np.ndarray, np.ndarray, np.ndarray]: image = image.copy() if image.mode != 'RGB': image = image.convert('RGB') small_image = _image_resize(image, pixels) few_raw = np.asarray(small_image).reshape(-1, 3) kmeans = KMeans(n_clusters=n) kmeans.fit(few_raw) width, height = image.size raw = np.asarray(image).reshape(-1, 3) colors = kmeans.cluster_centers_.round().astype(np.uint8) prediction = kmeans.predict(raw) new_data = colors[prediction].reshape((height, width, 3)) new_image = Image.fromarray(new_data, mode='RGB') cids, counts = np.unique(prediction, return_counts=True) counts = np.asarray(list(map(lambda x: x[1], sorted(zip(cids, counts))))) return new_image, colors, counts, prediction.reshape((height, width)) def main_func(image: Image.Image, n: int, pixels: int, fixed_width: bool, width: int): if fixed_width: _width, _height = image.size r = width / _width new_width, new_height = int(round(_width * r)), int(round(_height * r)) image = image.resize((new_width, new_height)) new_image, colors, counts, predictions = get_main_colors(image, n, pixels) table = pd.DataFrame({ 'r': colors[:, 0], 'g': colors[:, 1], 'b': colors[:, 2], 'count': counts, }) table['ratio'] = table['count'] / table['count'].sum() hexes = [] for r, g, b in zip(table['r'], table['g'], table['b']): hexes.append(f'#{r:02x}{g:02x}{b:02x}') table['hex'] = hexes new_table = pd.DataFrame({ 'Hex': table['hex'], 'Pixels': table['count'], 'Ratio': table['ratio'], 'Red': table['r'], 'Green': table['g'], 'Blue': table['b'], }).sort_values('Pixels', ascending=False) return new_image, new_table if __name__ == '__main__': pd.set_option("display.precision", 3) with gr.Blocks() as demo: with gr.Row(): with gr.Column(): ch_image = gr.Image(type='pil', label='Original Image') with gr.Row(): ch_clusters = gr.Slider(value=8, minimum=2, maximum=256, step=2, label='Clusters') ch_pixels = gr.Slider(value=100000, minimum=10000, maximum=1000000, step=10000, label='Pixels for Clustering') ch_fixed_width = gr.Checkbox(value=True, label='Width Fixed') ch_width = gr.Slider(value=200, minimum=12, maximum=2048, label='Width') ch_submit = gr.Button(value='Submit', variant='primary') with gr.Column(): with gr.Tabs(): with gr.Tab('Output Image'): ch_output = gr.Image(type='pil', label='Output Image') with gr.Tab('Color Map'): ch_color_map = gr.Dataframe( headers=['Hex', 'Pixels', 'Ratio', 'Red', 'Green', 'Blue'], label='Color Map' ) ch_submit.click( main_func, inputs=[ch_image, ch_clusters, ch_pixels, ch_fixed_width, ch_width], outputs=[ch_output, ch_color_map], ) demo.queue(os.cpu_count()).launch()