import streamlit as st from PIL import Image from modInference import main import numpy as np import math import numpy as np from skimage import transform import os from keras.models import Model from keras.applications.vgg16 import VGG16, preprocess_input from keras.layers import Dense, Dropout, Flatten import numpy as np import torch from models.create_fasterrcnn_model import create_model st.set_page_config(layout="wide") st.markdown("") showImg = Image.open('3dots.jpg') ogInp = Image.open('3dots.jpg') showImg = showImg.resize((100, 100)) ogInp = ogInp.resize((100, 100)) cellImgs = [] st.title('MicroScan In Action!') st.subheader("Enter an image of a thin blood smear. Preview the image and run the application. This program was developed by Anish Pallod =). Disclaimer: this is currently a prototype and shouldn't be used for medical use. ") input, outputIm, col, col1, col2, col3, col4, col5, col6, col7 = st.columns((5,5,1,1,1,1,1,1,1,1)) def load_model(): NUM_CLASSES = 2 CLASSES = ['__background__', 'Cell'] DEVICE = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') print(DEVICE) checkpoint = torch.load("best_model.pth", map_location=DEVICE) NUM_CLASSES = checkpoint['config']['NC'] CLASSES = checkpoint['config']['CLASSES'] build_model = create_model[checkpoint['model_name']] model = build_model(num_classes=NUM_CLASSES, coco_model=False) model.load_state_dict(checkpoint['model_state_dict']) model.to(DEVICE).eval() COLORS = np.random.uniform(0, 255, size=(len(CLASSES), 3)) conv_base = VGG16(include_top=False, weights='imagenet', input_shape=(200,200,3)) if 2 > 0: for layer in conv_base.layers[:-2]: layer.trainable = False else: for layer in conv_base.layers: layer.trainable = False top_model = conv_base.output top_model = Flatten(name="flatten")(top_model) top_model = Dense(4096, activation='relu')(top_model) top_model = Dense(1048, activation='relu')(top_model) top_model = Dense(256, activation='relu')(top_model) top_model = Dense(128, activation='relu')(top_model) top_model = Dense(64, activation='relu')(top_model) top_model = Dropout(0.2)(top_model) output_layer = Dense(5, activation='softmax')(top_model) CNN = Model(inputs=conv_base.input, outputs=output_layer) CNN.load_weights("CNN.hdf5") return CNN, model CNN, model = load_model() with input: st.header("Input") imageInput = st.file_uploader("Enter an image of a thin blood smear.") if st.button("Run"): if imageInput is not None: image = Image.open(imageInput) ogInp = image img_array = np.array(image) output, cellImgs = main(CNN, model, img_array) showImg = Image.fromarray(output) if st.button("Preview"): if imageInput is not None: image = Image.open(imageInput) ogInp = image st.write("-" * 34) st.header("How it looks:") st.image(ogInp) else: st.write("-" * 34) st.header("How it looks:") st.image(ogInp) with outputIm: st.header("General Output") st.image(showImg) total = len(cellImgs) print(cellImgs) barrier = [] for k in range(1, 8): barrier.append(math.floor(total/7) * k) leftOver = total % 7 for k in range(leftOver): barrier[k] += 1 print(barrier) st.markdown(""" """,unsafe_allow_html=True) total = len(cellImgs) print(cellImgs) barrier = [] for k in range(1, 8): barrier.append(math.floor(total/7) * k) leftOver = total % 7 for k in range(leftOver): barrier[k] += 1 print(barrier) with col1: for x in cellImgs[0:barrier[0]]: st.write(x[1]) st.image(x[0]) with col2: for x in cellImgs[barrier[0]: barrier[1]]: st.write(x[1]) st.image(x[0]) with col3: for x in cellImgs[barrier[1]: barrier[2]]: st.write(x[1]) st.image(x[0]) with col4: for x in cellImgs[barrier [2]: barrier[3]]: st.write(x[1]) st.image(x[0]) with col5: for x in cellImgs[barrier[3]: barrier[4]]: st.write(x[1]) st.image(x[0]) with col6: for x in cellImgs[barrier[4]: barrier[5]]: st.write(x[1]) st.image(x[0]) with col7: for x in cellImgs[barrier[5]: barrier[6]]: st.write(x[1]) st.image(x[0])