app.py

import streamlit as st

from fibsem.segmentation.model import load_model
from fibsem.structures import FibsemImage
import matplotlib.pyplot as plt
import glob

from PIL import Image
import numpy as np

st.set_page_config(layout="wide")


st.title("Autolamella Demo")
st.write("This is a demo space of the Autolamella models. Upload FIBSEM (tiff) image, and select a model to segment them.")

# get filenames
filenames = st.sidebar.file_uploader("Upload an image", type=["tiff", "tif"], accept_multiple_files=True)

# get model
checkpoint = st.sidebar.selectbox(
    "Select a model checkpoint",
    [
     "autolamella-mega-20240107.pt",
     "autolamella-waffle-20240107.pt", 
     "autolamella-serial-liftout-20240107.pt",
     ],)

st.sidebar.header("Available Models")
st.sidebar.write("""The following models are available for inference. They are trained on different datasets, and may perform differently on different samples.""")

# write a markdown list, listing each of the models
st.sidebar.write("""
* autolamella-waffle*
* autolamella-serial-liftout*
* autolamella-mega*""")         
st.sidebar.write("If you have a new sample, try all of them and see which one works best.""")

# get default data from path
if len(filenames) == 0:
    st.write("No files uploaded, using default data")
    # get all tiff files in current directory
    filenames = sorted(glob.glob("example/*.tif"))


st.header(f"Segmentation Results")

# load model
model = load_model(checkpoint)

if filenames:
    
    cols = st.columns(4)

    for i, fname in enumerate(filenames):
        
        col_id = i % 4
        cols[col_id].write(f"#### File: {fname} ({i+1}/{len(filenames)})")
        cols[col_id].write(f"Running inference... ")
        # load image, segment, and save
        image = FibsemImage.load(fname)

        # resize to 1536 x 1024
        image.data = np.asarray(Image.fromarray(image.data).resize((1536, 1024)))

        mask = model.inference(image.data, rgb=True)

        # plot
        fig = plt.figure(figsize=(10, 10))
        plt.imshow(image.data, cmap="gray")
        plt.imshow(mask, alpha=0.5)
        plt.axis("off")


        cols[col_id].pyplot(fig, use_container_width=True)