app.py

import gradio as gr
from transformers import pipeline, CLIPProcessor, CLIPModel
from PIL import Image
import torch
import openai
import base64
import io

# ── 1. DEIN MODELL (von Hugging Face) ──────────────────────────
MY_MODEL_ID = "DEIN-USERNAME/DEIN-MODELL"  # ← anpassen!
my_classifier = pipeline("image-classification", model=MY_MODEL_ID)

# ── 2. CLIP (Open-Source) ──────────────────────────────────────
clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")

# Deine Klassen (anpassen!)
LABELS = ["cat", "dog", "bird"]  # ← deine eigenen Klassen

def predict_my_model(image):
    results = my_classifier(image)
    return {r["label"]: r["score"] for r in results}

def predict_clip(image):
    inputs = clip_processor(
        text=LABELS, images=image, return_tensors="pt", padding=True
    )
    with torch.no_grad():
        outputs = clip_model(**inputs)
    probs = outputs.logits_per_image.softmax(dim=1)[0]
    return {label: float(prob) for label, prob in zip(LABELS, probs)}

def predict_openai(image):
    client = openai.OpenAI(api_key=openai.api_key)
    
    # Bild zu Base64 konvertieren
    buf = io.BytesIO()
    image.save(buf, format="JPEG")
    b64 = base64.b64encode(buf.getvalue()).decode()
    
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "user",
            "content": [
                {"type": "image_url",
                 "image_url": {"url": f"data:image/jpeg;base64,{b64}"}},
                {"type": "text",
                 "text": f"Classify this image as one of: {LABELS}. "
                         f"Return only a JSON like: {{\"label\": score, ...}} "
                         f"where scores sum to 1."}
            ]
        }],
        max_tokens=100
    )
    import json
    return json.loads(response.choices[0].message.content)

def classify_all(image):
    r1 = predict_my_model(image)
    r2 = predict_clip(image)
    r3 = predict_openai(image)
    return r1, r2, r3

# ── Beispielbilder ─────────────────────────────────────────────
examples = ["example1.jpg", "example2.jpg"]  # ← eigene Bilder

# ── Gradio Interface ───────────────────────────────────────────
with gr.Blocks(title="Image Classification Comparison") as demo:
    gr.Markdown("# 🖼️ Image Classification – Model Comparison")
    gr.Markdown("Compare your custom model, CLIP, and GPT-4o Vision.")
    
    with gr.Row():
        img_input = gr.Image(type="pil", label="Upload Image")
    
    btn = gr.Button("Classify!", variant="primary")
    
    with gr.Row():
        out1 = gr.Label(label="🏷️ My Model")
        out2 = gr.Label(label="🔍 CLIP")
        out3 = gr.Label(label="🤖 GPT-4o")
    
    btn.click(classify_all, inputs=img_input, outputs=[out1, out2, out3])
    gr.Examples(examples=examples, inputs=img_input)

demo.launch()