app.py

from fastai.vision.all import *
import gradio as gr  # Import Gradio directly
import timm

# Load your model
learn = load_learner('model.pkl')
categories = learn.dls.vocab

# Define your prediction function
def classify_image(img):
    pred, idx, probs = learn.predict(img)
    # Align with Gradio's formatting for outputs
    return {category: prob for category, prob in zip(categories, probs)}

# Create the Gradio interface
image = gr.Image()  # Use gr.Image directly for input
label = gr.Label(num_top_classes=len(categories))  # Adjust for multi-class output

interface = gr.Interface(
    fn=classify_image,  # Reference the function directly
    inputs=image,
    outputs=label
)

interface.launch()