app.py

import gradio as gr
from PIL import Image
from skimage import color, feature
from skimage.segmentation import slic, mark_boundaries
from skimage.filters import threshold_otsu
import numpy as np
import cv2
import os

# 驗證輸入的圖片是否為有效的 NumPy 陣列格式
def validate_image(image):
    if not isinstance(image, np.ndarray):
        raise ValueError("Input is not a valid NumPy array.")
    
    if image.ndim == 3 and image.shape[2] == 4:  # RGBA to RGB
        image = image[:, :, :3]
    
    if image.ndim == 2:  # Grayscale to RGB
        image = np.stack([image] * 3, axis=-1)
    elif image.ndim == 3 and image.shape[2] != 3:
        raise ValueError("Image must have 3 channels (RGB).")
    
    return image.astype(np.uint8)

# 邊緣檢測函數
def edge_detection(image, low_threshold, high_threshold):
    gray = color.rgb2gray(image)
    edges = feature.canny(gray, sigma=low_threshold)
    return (edges * 255).astype(np.uint8)

# 影像分割函數
def image_segmentation(image, num_segments):
    segments = slic(image, n_segments=num_segments, start_label=1)
    segmented_image = mark_boundaries(image, segments)
    return (segmented_image * 255).astype(np.uint8)

# 修復影像函數（圖像修補）
def image_inpainting(image, mask=None):
    if mask is None:
        gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
        _, mask = cv2.threshold(gray, threshold_otsu(gray), 255, cv2.THRESH_BINARY_INV)
        mask = cv2.dilate(mask, None, iterations=3)
    
    if mask.ndim == 3:
        mask = cv2.cvtColor(mask, cv2.COLOR_RGB2GRAY)
    
    inpainted = cv2.inpaint(image, mask, inpaintRadius=3, flags=cv2.INPAINT_TELEA)
    return inpainted

# 主處理函數，用來根據選擇的任務進行處理
def process_image(task, image, mask, param1, param2):
    try:
        # 驗證輸入的圖像
        image = validate_image(image)
        
        if task == "Edge Detection":
            result = edge_detection(image, param1, param2)
        elif task == "Image Segmentation":
            result = image_segmentation(image, int(param1))
        elif task == "Image Inpainting":
            result = image_inpainting(image, mask)
        else:
            raise ValueError(f"Unknown task: {task}")
        
        result = np.clip(result, 0, 255).astype(np.uint8)
        
        # 確保返回 PIL Image 格式
        return Image.fromarray(result)  # 返回 PIL Image 格式的圖像
        
    except Exception as e:
        print(f"錯誤: {e}")
        return np.zeros((100, 100, 3), dtype=np.uint8)

# 生成範例資料，強化對圖像的檢查和處理
def generate_grouped_examples(task_names, examples_dir):
    examples = []
    images = sorted([f for f in os.listdir(examples_dir) if f.endswith((".png", ".jpg", ".jpeg"))])

    # 生成一個全黑的 mask 當作 placeholder
    black_mask = Image.fromarray(np.zeros((100, 100, 3), dtype=np.uint8))

    # for file_name in images:
    #     file_path = os.path.join(examples_dir, file_name)
    #     try:
    #         image = Image.open(file_path).convert("RGB")
    #         image_array = np.array(image)

    #         # 確保範例圖像轉為 PIL.Image 格式
    #         image_pil = Image.fromarray(image_array)

    #         # for task in task_names:
    #         #     # 確保範例圖像正確
    #         #     examples.append([task, image_pil, None, 2, 5])  # 傳入 PIL 圖像
    #         for task in task_names:
    #             if task == "Image Inpainting":
    #                 # 給一個 dummy mask
    #                 examples.append([task, image_pil, black_mask, 2, 5])
    #             else:
    #                 # 對非 Inpainting 任務，可直接給 mask=None 或改給 black_mask，看哪種運行正常
    #                 # 但通常給 None 就會引發現在的問題，所以建議也用黑 mask
    #                 examples.append([task, image_pil, black_mask, 2, 5])
    #     except Exception as e:
    #         print(f"Error loading example {file_name}: {e}")
    
    # return examples
    for i, file_name in enumerate(images):
        file_path = os.path.join(examples_dir, file_name)
        try:
            image = Image.open(file_path).convert("RGB")
            image_array = np.array(image)
            image_pil = Image.fromarray(image_array)

            # 根據 i 決定前三張或其他張如何設定 mask
            for task in task_names:
                if i < 2:
                    # 前三張圖片：第三個參數 (mask) = None
                    examples.append([task, image_pil, None, 2, 5])
                else:
                    # 之後的圖片：如果是 Inpainting，就給 black_mask；否則也可給黑 mask
                    if task == "Image Inpainting":
                        examples.append([task, image_pil, black_mask, 2, 5])
                    else:
                        examples.append([task, image_pil, black_mask, 2, 5])

        except Exception as e:
            print(f"Error loading example {file_name}: {e}")
    
    return examples


tasks = ["Edge Detection", "Image Segmentation", "Image Inpainting"]
examples_dir = "examples"  # 假設範例圖像在此目錄下
examples = generate_grouped_examples(tasks, examples_dir)

# Gradio 介面設定
interface = gr.Interface(
    fn=process_image,
    inputs=[  
        gr.Dropdown(choices=tasks, label="Task"),  
        gr.Image(type="numpy", label="Input Image"),  
        gr.Image(type="numpy", label="Mask (Optional), Note: This is for Image Inpainting function"),  
        gr.Slider(1, 10, value=1, label="Parameter 1 (e.g., Edge Sigma, Num Segments)"),  
        gr.Slider(0, 10, value=1, label="Parameter 2 (e.g., High Threshold)")  
    ],
    outputs=gr.Image(type="pil"),  # 返回 PIL 格式的圖像
    examples=examples,  
    title="Computer Vision Web App",
    description="Perform Edge Detection, Image Segmentation, or Inpainting on uploaded images."
)

if __name__ == "__main__":
    interface.launch()