app.py

from transformers import BlipProcessor, BlipForConditionalGeneration
from PIL import Image, ImageEnhance, ImageOps, ImageFilter
from translate import Translator


from IPython.display import Image
try:
  filename = take_photo()
  print('Saved to {}'.format(filename))

  # Show the image which was just taken.
  display(Image(filename))
except Exception as err:
  # Errors will be thrown if the user does not have a webcam or if they do not
  # grant the page permission to access it.
  print(str(err))
    
from transformers import BlipProcessor, BlipForConditionalGeneration
from PIL import Image, ImageEnhance, ImageOps, ImageFilter
from translate import Translator

# 載入BLIP模型和處理器
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")

def generate_caption(image, language="中文"):
    try:
        # 將圖片轉換為模型可以理解的格式
        inputs = processor(image, return_tensors="pt")

        # 生成更具體的描述，透過設置 prompt
        prompt = (
            "Describe the image in detail, including objects, actions, "
            "colors, and overall context, to make the description more complete."
        )
        out = model.generate(
            **inputs,
            max_length=100,  # 增加描述長度
            num_beams=7,     # 提高生成的多樣性
            no_repeat_ngram_size=3,  # 降低重複率
            temperature=200,         # 增加生成描述的靈活性
            top_k=50,
            top_p=0.95
        )
        caption = processor.decode(out[0], skip_special_tokens=True).strip()

        # 翻譯為指定語言
        if language != "English":  # 如果不是英文，才進行翻譯
            lang_code_map = {
                "中文": "zh-tw",
                "法文": "fr",
                "德文": "de",
                "西班牙文": "es",
                "日文": "ja",
                "阿拉伯文": "ar"
            }
            translator = Translator(to_lang=lang_code_map[language])
            caption = translator.translate(caption)

        return caption
    except Exception as e:
        return f"描述生成失敗: {str(e)}"

def change_style(image, style):
    if style == "黑白":
        image = image.convert("L")
    elif style == "模糊":
        image = image.filter(ImageFilter.BLUR)
    elif style == "銳化":
        image = image.filter(ImageFilter.SHARPEN)
    elif style == "邊緣增強":
        image = image.filter(ImageFilter.EDGE_ENHANCE)
    elif style == "反轉顏色":
        image = ImageOps.invert(image.convert("RGB"))
    elif style == "懷舊":
        sepia_filter = ImageEnhance.Color(image.convert("RGB"))
        image = sepia_filter.enhance(0.3)
    return image

def process_image(image, style, language):
    caption = generate_caption(image, language)
    styled_image = change_style(image, style)
    return caption, styled_image

# 設定Gradio介面
import gradio as gr
interface = gr.Interface(
    fn=process_image,
    inputs=[
        gr.Image(type="pil", label="上傳圖片或使用攝像頭"),
        gr.Radio(["原始", "黑白", "模糊", "銳化", "邊緣增強", "反轉顏色", "懷舊"], label="選擇風格"),
        gr.Radio(["中文", "English", "法文", "德文", "西班牙文", "日文", "阿拉伯文"], label="選擇語言")
    ],
    outputs=[
        gr.Textbox(label="圖片描述"),
        gr.Image(type="pil", label="變換畫風後的圖像")
    ],
    title="圖片描述與畫風變換（更具體描述）"
)

# 啟動介面
interface.launch()