app.py

"""
This module provides an interface for classifying images using the ResNet-18 model.
The interface allows users to upload an image and receive the top 3 predicted labels.
"""

import gradio as gr
import spaces
import torch
from transformers import AutoImageProcessor, AutoModelForImageClassification
from datasets import load_dataset
from PIL import Image

# Load dataset and get test image
dataset = load_dataset("huggingface/cats-image", trust_remote_code=True)
test_image = dataset["test"]["image"][0]

# Initialize the image processor and model
image_processor = AutoImageProcessor.from_pretrained("microsoft/resnet-18")
model = AutoModelForImageClassification.from_pretrained("microsoft/resnet-18").to("cuda")

@spaces.GPU
def predict(image: Image, top_k: int = 3) -> dict:
    """
    Predicts the top 'top_k' labels for an image using the ResNet-18 model.

    Args:
        image (Image): The input image as a PIL Image object.
        top_k (int): The number of top predictions to return.

    Returns:
        dict: A dictionary with the top 'top_k' labels and their probabilities.
    """
    inputs = image_processor(image, return_tensors="pt").to("cuda")
    with torch.no_grad():
        logits = model(**inputs).logits

    # Apply softmax to logits to get probabilities
    probabilities = torch.softmax(logits, dim=-1)

    # Get the top 'top_k' probabilities and their corresponding indices
    top_k_probs, top_k_indices = torch.topk(input=probabilities, k=top_k, dim=-1)

    # Map the indices to labels and probabilities
    predicted_labels = {
        model.config.id2label[idx.item()]: prob.item()
        for idx, prob in zip(top_k_indices[0], top_k_probs[0])
    }
    
    return predicted_labels

# Define the Gradio interface
demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=3),
    title="Classifying Images with ResNet-18",
    description="Upload an image to predict the top 3 labels.",
    examples=[test_image]
)

# Launch the Gradio interface
demo.launch()