app.py

import streamlit as st
import sahi.utils.file
from PIL import Image
from sahi import AutoDetectionModel
from utils import sahi_yolov8m_inference
from streamlit_image_comparison import image_comparison
from ultralyticsplus.hf_utils import download_from_hub

IMAGE_TO_URL = {
    'factory_pid.png' : 'https://d1afc1j4569hs1.cloudfront.net/factory-pid.png',
    'plant_pid.png' : 'https://d1afc1j4569hs1.cloudfront.net/plant-pid.png',
    'processing_pid.png' : 'https://d1afc1j4569hs1.cloudfront.net/processing-pid.png',
    'prediction_visual.png' : 'https://d1afc1j4569hs1.cloudfront.net/prediction_visual.png'
    }

st.set_page_config(
    page_title="P&ID Object Detection",
    layout="wide",
    initial_sidebar_state="expanded"
    )

st.title('P&ID Object Detection')
st.subheader(' Identify valves and pumps with deep learning model ', divider='rainbow')
st.caption('Developed by Deep Drawings Co.')

@st.cache_resource(show_spinner=False)
def get_model(postprocess_match_threshold):
    yolov8_model_path = download_from_hub('DanielCerda/pid_yolov8')
    detection_model = AutoDetectionModel.from_pretrained(
        model_type='yolov8',
        model_path=yolov8_model_path,
        confidence_threshold=postprocess_match_threshold,
        device="cpu",
    )
    return detection_model
    
@st.cache_data(show_spinner=False)
def download_comparison_images():
    sahi.utils.file.download_from_url(
        'https://d1afc1j4569hs1.cloudfront.net/plant-pid.png',
        'plant_pid.png',
    )
    sahi.utils.file.download_from_url(
        'https://d1afc1j4569hs1.cloudfront.net/prediction_visual.png',
        'prediction_visual.png',
    )

download_comparison_images()

if "output_1" not in st.session_state:
    img_1 = Image.open('plant_pid.png')
    st.session_state["output_1"] = img_1.resize((4960,3508))

if "output_2" not in st.session_state:
    img_2 = Image.open('prediction_visual.png')
    st.session_state["output_2"] = img_2.resize((4960,3508))

col1, col2, col3 = st.columns(3, gap='medium')
with col1:
    with st.expander('How to use it'):
        st.markdown(
        '''
        1) Upload or select any example diagram  👆🏻
        2) Set model parameters 📈
        3) Press to perform inference   🚀
        4) Visualize model predictions  🔎
        '''
        )   

st.write('##')
   
col1, col2, col3 = st.columns(3, gap='large')
with col1:
    st.markdown('##### Set Input Image')
    # set input image by upload
    image_file = st.file_uploader(
        'Upload your P&ID', type = ['jpg','jpeg','png']
    )
    # set input images from examples
    def radio_func(option):
        option_to_id = {
            'factory_pid.png' : 'A',
            'plant_pid.png' : 'B',
            'processing_pid.png' : 'C',
        }
        return option_to_id[option]
    radio = st.radio(
        'Select from the following examples',
        options = ['factory_pid.png', 'plant_pid.png', 'processing_pid.png'],
        format_func = radio_func,
    )
with col2:
    # visualize input image
    if image_file is not None:
        image = Image.open(image_file)
    else:
        image = sahi.utils.cv.read_image_as_pil(IMAGE_TO_URL[radio])
    st.markdown('##### Preview')
    with st.container(border = True):
        st.image(image, use_column_width = True)
        
with col3:
    # set SAHI parameters
    st.markdown('##### Set model parameters')
    slice_number = st.select_slider(
        'Slices per Image', 
        options = [
            '1',
            '4',
            '16',
            '64',            
        ],
    )
    overlap_ratio = st.slider(
        label = 'Slicing Overlap Ratio', 
        min_value=0.0, 
        max_value=0.5, 
        value=0.1, 
        step=0.1
    )
    postprocess_match_threshold = st.slider(
        label = 'Confidence Threshold',
        min_value = 0.0,
        max_value = 1.0,
        value = 0.8,
        step = 0.1
    )

st.write('##')

col1, col2, col3 = st.columns([3, 1, 3])
with col2:
    submit = st.button("🚀 Perform Prediction")
    
if submit:
    # perform prediction
    with st.spinner(text="Downloading model weights ... "):
        detection_model = get_model(postprocess_match_threshold)
        
    slice_size = int(4960/(slice_number**0.5))
    image_size = 4960

    with st.spinner(text="Performing prediction ... "):
        output = sahi_yolov8m_inference(
            image,
            detection_model,
            image_size=image_size,
            slice_height=slice_size,
            slice_width=slice_size,
            overlap_height_ratio=overlap_ratio,
            overlap_width_ratio=overlap_ratio,
        )

    st.session_state["output_1"] = image
    st.session_state["output_2"] = output 

st.write('##')

col1, col2, col3 = st.columns([3, 1, 1], gap='small')
with col1:
    st.markdown(f"#### Object Detection Result")
    with st.container(border = True):
        static_component = image_comparison(
        img1=st.session_state["output_1"],
        img2=st.session_state["output_2"],
        label1='Raw Diagram',
        label2='Inference Prediction',
        width=768,
        starting_position=50,
        show_labels=True,
        make_responsive=True,
        in_memory=True,
        )