test_accuracy.py

import os
import requests
import io
from PIL import Image
from app import get_image_detector

def download_image(url):
    response = requests.get(url, headers={"User-Agent": "Mozilla/5.0"})
    response.raise_for_status()
    return Image.open(io.BytesIO(response.content)).convert("RGB")

def run_tests():
    detector = get_image_detector()
    
    test_cases = [
        {
            "name": "Real Image - Dog",
            "url": "https://images.unsplash.com/photo-1543466835-00a7907e9de1?q=80&w=800&auto=format&fit=crop",
            "expected": "REAL"
        },
        {
            "name": "Fake Image - AI Face",
            "url": "https://thispersondoesnotexist.com/",
            "expected": "FAKE"
        },
        {
             "name": "Real Image - Architecture",
             "url": "https://upload.wikimedia.org/wikipedia/commons/4/47/New_york_times_square-terabass.jpg",
             "expected": "REAL"
        },
        {
             "name": "Real Image - People",
             "url": "https://images.unsplash.com/photo-1517841905240-472988babdf9?ixlib=rb-4.0.3&q=80&w=800&auto=format&fit=crop",
             "expected": "REAL"
        }
    ]
    
    print("starting accuracy tests...")
    correct = 0
    for case in test_cases:
        print(f"\nTesting: {case['name']} (Expected: {case['expected']})")
        print(f"URL: {case['url']}")
        try:
            img = download_image(case['url'])
            result, _ = detector.predict_with_visuals(img, include_gradcam=False, include_fft=False, include_result_card=False)
            label = result['label']
            
            print(f"Result: {label}")
            print(f"Fake Prob: {result['fake_prob']:.4f}")
            print(f"Real Prob: {result['real_prob']:.4f}")
            
            for model, score in result['scores'].items():
                print(f"  {model}: {score:.4f}")
                
            if (label == case['expected']) or (label == "FAKE" and case['expected'] == "FAKE") or (label == "REAL" and case['expected'] == "REAL"):
                print("[Correct]")
                correct += 1
            else:
                print("[Incorrect]")
                
        except Exception as e:
            print(f"[Error] testing {case['name']}: {e}")
            
    print(f"\nAccuracy: {correct}/{len(test_cases)} ({correct/len(test_cases)*100:.1f}%)")

if __name__ == "__main__":
    run_tests()