jedzqg/is_anime_or_real

This is based on the ResNet18 model and is used to determine whether an image is anime or real-life photography. The accuracy has limitations. To use the model, you can refer to the following code:

  import torch
  import torch.nn as nn
  import torch.nn.functional as F
  from torchvision import transforms
  from PIL import Image
  import os
  from torchvision import models
  
  # Set up device configuration to use GPU if available
  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
  
  # Load pre-trained ResNet18 model and modify the final layer for 2-class classification
  model = models.resnet18()
  model.fc = nn.Linear(model.fc.in_features, 2)
  model.load_state_dict(torch.load('resnet18_anime_real.pth', map_location=device))
  model.to(device)
  model.eval()
  
  # Image preprocessing pipeline matching the training setup (uses ImageNet mean/std)
  transform = transforms.Compose([
      transforms.Resize((224, 224)),  # Input size for ResNet
      transforms.ToTensor(),
      transforms.Normalize(mean=[0.485, 0.456, 0.406],
                           std=[0.229, 0.224, 0.225])
  ])
  
  # Define inference function to classify images
  def predict_image(img_path):
      image = Image.open(img_path).convert('RGB')
      image = transform(image).unsqueeze(0).to(device)  # Add batch dimension [1, 3, H, W]
  
      with torch.no_grad():  # Disable gradient calculation for inference
          output = model(image)
          predicted = torch.argmax(output, 1).item()
          # Print and return prediction results
          if predicted == 0:
              print("✨ Prediction: This is an anime-style image! (≧∇≦)ﾉ")
              return 0
          else:
              print("📸 Prediction: This is a real photo! (*≧ω≦)")
              return 1
  
  # Main execution for testing the model on image datasets
  if __name__ == '__main__':
      # Evaluate on anime images
      anime_correct = total_anime = 0
      for filename in sorted(os.listdir('./test_anime')):
          if filename.lower().endswith(('.jpg', '.jpeg', '.png', '.bmp', '.webp')):
              test_image_path = "./test_anime/" + filename
              result = predict_image(test_image_path)
              anime_correct += 1 if result == 0 else 0
              total_anime += 1
      
      print("Starting evaluation on real photos")
      # Evaluate on real photos
      real_correct = total_real = 0
      for filename in sorted(os.listdir('./test_real')):
          if filename.lower().endswith(('.jpg', '.jpeg', '.png', '.bmp', '.webp')):
              test_image_path = "./test_real/" + filename
              result = predict_image(test_image_path)
              real_correct += 1 if result == 1 else 0
              total_real += 1
      
      # Calculate and print accuracy metrics
      print(f"Anime image accuracy: {anime_correct}/{total_anime} = {anime_correct/total_anime:.2%}")
      print(f"Real photo accuracy: {real_correct}/{total_real} = {real_correct/total_real:.2%}")