app.py

import streamlit as st
import torch
from transformers import AutoProcessor, AutoModelForVision2Seq
from PIL import Image

device = "cuda" if torch.cuda.is_available() else "cpu"

# Load model and processor
@st.cache_resource
def load_model():
    model_name = "tjoab/latex_finetuned"
    processor = AutoProcessor.from_pretrained(model_name)
    model = AutoModelForVision2Seq.from_pretrained(model_name).to(device)
    return processor, model



processor, model = load_model()

st.title("LaTeX Image to Text Converter")
st.write("Upload an image containing a handwritten or printed math expression, and get the LaTeX code.")

# TODO: Add .png support (doesnt work as is with PIL Image.open())
uploaded_file = st.file_uploader("Choose an image...", type=["jpg", "jpeg"])

if uploaded_file:
    image = Image.open(uploaded_file).convert("RGB")
    st.image(image, caption="Uploaded Image", use_column_width=True)

    # Preprocess image as model expects, then run inference
    with st.spinner("Processing..."):
        preproc_image = processor.image_processor(image, return_tensors="pt").pixel_values
        preproc_image = preproc_image.to(device)

        pred_ids = model.generate(preproc_image, max_length=128)
        latex_pred = processor.batch_decode(pred_ids, skip_special_tokens=True)[0]

    st.subheader("Predicted LaTeX Code:")
    st.code(latex_pred, language="latex")