import cv2 import numpy as np from io import BytesIO from PIL import Image from fastapi import FastAPI, File, UploadFile from keras.models import load_model model = load_model("model-dr.h5") def preprocess(img): cv2_img = np.array(img) image = cv2.cvtColor(cv2_img, cv2.COLOR_RGB2BGR) image = cv2.resize(image,(400,400)) image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) kopya = image.copy() kopya = cv2.cvtColor(kopya, cv2.COLOR_RGB2GRAY) blur = cv2.GaussianBlur(kopya,(5,5),0) thresh = cv2.threshold(blur,10,255, cv2.THRESH_BINARY)[1] kontur = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) kontur = kontur[0][0] kontur = kontur[:,0,:] x1 = tuple(kontur[kontur[:,0].argmin()])[0] y1 = tuple(kontur[kontur[:,1].argmin()])[1] x2 = tuple(kontur[kontur[:,0].argmax()])[0] y2 = tuple(kontur[kontur[:,1].argmax()])[1] x = int(x2-x1)*4//50 y = int(y2-y1)*5//50 kopya2 = image.copy() if x2-x1 >100 and y2-y1> 100: kopya2 = kopya2[y1+y : y2-y , x1+x : x2-x] kopya2 = cv2.resize(kopya2,(400,400)) lab = cv2.cvtColor(kopya2, cv2.COLOR_RGB2LAB) l,a,b = cv2.split(lab) clahe = cv2.createCLAHE(clipLimit=5.0,tileGridSize=((8,8))) cl = clahe.apply(l) limg = cv2.merge((cl,a,b)) son = cv2.cvtColor(limg, cv2.COLOR_LAB2RGB) med_son = cv2.medianBlur(son, 3) arka_plan = cv2.medianBlur(son, 37) maske = cv2.addWeighted(med_son,1,arka_plan,-1,255) son_img = cv2.bitwise_and(maske,med_son) return son_img app = FastAPI() @app.get('/') async def index(): return "Welcome DR Prediction" def read_imagefile(file) -> Image.Image: image = Image.open(BytesIO(file)) return image @app.post("/predict") async def predict_api(file: UploadFile = File(...)): extension = file.filename.split(".")[-1] in ("jpg", "jpeg", "png") if not extension: return "Image must be jpg or png format!" image = read_imagefile(await file.read()) image = preprocess(image) prediction = model.predict(np.expand_dims(image,axis=0))[0] no_dr = format(prediction[0]*100,".2f") mild = format(prediction[1]*100,".2f") moderate = format(prediction[2]*100,".2f") severe = format(prediction[3]*100,".2f") pdr = format(prediction[4]*100,".2f") return {"no_dr":no_dr, "mild":mild, "moderate":moderate, "severe":severe, "pdr":pdr, }