example_inference.py

#!/usr/bin/env python
# -*- coding:utf-8 -*-

import torch
from PIL import Image
from transformers import AutoImageProcessor, AutoModelForImageClassification
import os

def main():
    model_dir = os.path.dirname(os.path.abspath(__file__))
    
    print("Loading model and processor...")
    processor = AutoImageProcessor.from_pretrained(model_dir)
    model = AutoModelForImageClassification.from_pretrained(model_dir)
    model.eval()
    
    test_image_path = os.path.join(model_dir, "assets", "164.jpg")
    if not os.path.exists(test_image_path):
        print(f"Test image not found: {test_image_path}")
        print("Please add a test image to assets/ folder")
        return
    
    print(f"Loading image: {test_image_path}")
    image = Image.open(test_image_path).convert("RGB")
    
    print("Processing image...")
    inputs = processor(images=image, return_tensors="pt")
    
    print("Running inference...")
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        probs = torch.nn.functional.softmax(logits, dim=-1)
        
        top5 = torch.topk(probs[0], k=5)
        
        print("\nTop 5 Predictions:")
        print("-" * 60)
        for i, (prob, idx) in enumerate(zip(top5.values, top5.indices)):
            class_idx = idx.item()
            label = model.config.id2label[str(class_idx)]
            confidence = prob.item() * 100
            print(f"{i+1}. {label} | Confidence: {confidence:.2f}%")

if __name__ == "__main__":
    main()