app.py

import gradio as gr
from transformers import pipeline, AutoFeatureExtractor, AutoModel
from PIL import Image
import torch
import numpy as np
import joblib
from huggingface_hub import hf_hub_download

# Load transformer feature extractor
feature_extractor = AutoFeatureExtractor.from_pretrained("google/vit-base-patch16-224")
model = AutoModel.from_pretrained("google/vit-base-patch16-224")

# Load your trained classifier (e.g., sklearn model)
model_path = hf_hub_download(repo_id="canadianjosieharrison/texture-logistic-model", filename="classifier.joblib")
classifier = joblib.load(model_path)  # Make sure this exists

# Create label map
id2label = {0:"metal", 1:"stone", 2:"wood"}

# Prediction function
def classify_image(img: Image.Image):
    # Step 1: Extract features from image
    inputs = feature_extractor(images=img, return_tensors="pt")
    with torch.no_grad():
        outputs = model(**inputs)
        embedding = outputs.last_hidden_state[:, 0, :].numpy()  # CLS token

    # Step 2: Predict using classifier
    prediction = classifier.predict(embedding)
    return f"Predicted Material: {id2label[prediction[0]]}"

# Gradio interface
interface = gr.Interface(
    fn=classify_image,
    inputs=gr.Image(type="pil"),
    outputs="text",
    title="Material texture image classifier",
    description="Upload a detail texture image of a material and the model will predict whether it's metal, stone, or wood.",
    examples = ["metal.PNG", "stone.PNG", "wood.PNG"],
)

if __name__ == "__main__":
    interface.launch(share=True)