app.py

import gradio as gr
import cv2
import numpy as np

def apply_filter(image, filter_type, intensity):
    # 강도를 0.0에서 1.0 사이로 정규화
    normalized_intensity = intensity / 100.0

    if filter_type == "Grayscale":
        return convert_to_grayscale(image)
    elif filter_type == "Soft Glow":
        gaussian = cv2.GaussianBlur(image, (15, 15), 0)
        soft_glow = cv2.addWeighted(image, 1 - normalized_intensity, gaussian, normalized_intensity, 0)
        return soft_glow
    elif filter_type == "Portrait Enhancer":
        # 부드러운 피부 표현을 위한 필터 조합
        # Step 1: Detail Enhance 사용 최소화
        enhanced = cv2.detailEnhance(image, sigma_s=5, sigma_r=0.05 + 0.05 * normalized_intensity)
        
        # Step 2: Gaussian Blur 적용 (적당한 강도로)
        blurred = cv2.GaussianBlur(enhanced, (15, 15), 0)
        
        # Step 3: Bilateral Filter 적용 (경계 보존하면서 부드럽게)
        smoothed = cv2.bilateralFilter(blurred, d=9, sigmaColor=75, sigmaSpace=75)
        
        return smoothed
    elif filter_type == "Warm Tone":
        increase_red = np.array([[1.0 + 0.2 * normalized_intensity, 0.0, 0.0],
                                 [0.0, 1.0, 0.0],
                                 [0.0, 0.0, 1.0 - 0.2 * normalized_intensity]])
        warm_image = cv2.transform(image, increase_red)
        return warm_image
    elif filter_type == "Cold Tone":
        increase_blue = np.array([[1.0 - 0.2 * normalized_intensity, 0.0, 0.0],
                                  [0.0, 1.0, 0.0],
                                  [0.0, 0.0, 1.0 + 0.2 * normalized_intensity]])
        cold_image = cv2.transform(image, increase_blue)
        return cold_image
    elif filter_type == "High-Key":
        high_key = cv2.convertScaleAbs(image, alpha=1.0 + 0.2 * normalized_intensity, beta=30)
        return high_key
    elif filter_type == "Low-Key":
        low_key = cv2.convertScaleAbs(image, alpha=1.0 - 0.3 * normalized_intensity, beta=-30)
        return low_key
    elif filter_type == "Haze":
        haze = cv2.addWeighted(image, 1.0 - 0.3 * normalized_intensity, np.full(image.shape, 255, dtype=np.uint8), 0.3 * normalized_intensity, 0)
        return haze
    else:
        return image

def convert_to_grayscale(image):
    gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    return gray_image

def convert_and_save(image, filter_type, intensity):
    filtered_image = apply_filter(image, filter_type, intensity)
    output_path = "output.jpg"
    cv2.imwrite(output_path, filtered_image)
    return filtered_image, output_path

iface = gr.Interface(
    fn=convert_and_save, 
    inputs=[
        "image", 
        gr.Radio(
            ["Grayscale", "Soft Glow", "Portrait Enhancer", "Warm Tone", "Cold Tone", "High-Key", "Low-Key", "Haze"], 
            label="필터 선택"
        ),
        gr.Slider(minimum=1, maximum=100, value=50, label="필터 강도")
    ], 
    outputs=["image", "file"], 
    title="이미지 필터 및 흑백 변환기",
    description="이미지를 업로드하고 필터와 강도를 선택하면, 변환된 이미지를 JPG 파일로 다운로드할 수 있습니다."
)

iface.launch()