app.py

import gradio as gr
import requests
from PIL import Image
from transformers import BlipProcessor, BlipForConditionalGeneration
import time

# Lade den Blip-Prozessor und das Modell
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")

def caption(img, min_len, max_len):
    # Öffne das Bild
    raw_image = Image.open(img).convert('RGB')
  
    # Verarbeite das Bild mit dem Blip-Prozessor
    inputs = processor(raw_image, return_tensors="pt")
    
    # Generiere eine Beschreibung mit dem Modell
    out = model.generate(**inputs, min_length=min_len, max_length=max_len)
    return processor.decode(out[0], skip_special_tokens=True)

def greet(img, min_len, max_len):
    start = time.time()
    result = caption(img, min_len, max_len)
    end = time.time()
    total_time = str(end - start)
    result = result + '\n' + total_time + ' seconds'
    return result

# Gradio Interface erstellen
iface = gr.Interface(
    fn=greet, 
    title='Blip Image Captioning Large', 
    description="[Salesforce/blip-image-captioning-base](https://huggingface.co/Salesforce/blip-image-captioning-base) Runs on CPU",
    inputs=[gr.Image(type='filepath', label='Image'), 
            gr.Slider(label='Minimum Length', minimum=1, maximum=1000, value=30), 
            gr.Slider(label='Maximum Length', minimum=1, maximum=1000, value=100)], 
    outputs=gr.Textbox(label='Caption'),
    theme=gr.themes.Base(primary_hue="teal", secondary_hue="teal", neutral_hue="slate"),
)

# API aktivieren, um sie von außen zu nutzen
iface.launch(share=True, allow_api=True)  # API aktivieren