labeller_smart.py

import time
import pymongo
import ssl
from PIL import Image
import numpy as np
import matplotlib.pyplot as plt
import secret
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator

client = pymongo.MongoClient(secret.connstring,
                             ssl=True,
                             ssl_cert_reqs=ssl.CERT_NONE) #connstring is a variable comming from secret.py containing mongo db connection string
db=client.test

model = tf.keras.models.load_model('model/digit_model6.h5')

#preprocessing functions applied on picture to be predicted
datagen=ImageDataGenerator(
    samplewise_center=True,
    samplewise_std_normalization=True,
)


def predict_class(im):
    prediction = model.predict(datagen.flow(im))
    return int(prediction.argmax())#convert numpy int64 to python int in order to be able to write to database later on


# select digit which does not contain label yet
cursor=db.gas_digit.find({
    'label': {
        '$not': {
            '$in': [
                0, 1, 2, 3, 4, 5, 6, 7, 8, 9
            ]
        }
    }
},{"image":1,"date":1,"label":1,"digit_position":1})

for document in cursor:
    if(document.get("label")==10):# I use label 10 for data to be deleted
        db.gas_digit.delete_one({"_id":document["_id"]})
        continue

    print(document["_id"],document["date"],document["image"][0:5],document['digit_position'])

    # reshape data so that they can be displayed as picture
    # 28 px x 28 px x 3 channels
    pic=np.array(document["image"]).reshape(28,28,3)
    id=document["_id"]

    #preprocess picture for display
    pic=(pic-pic.min())/(pic.max()-pic.min())#normalize
    pic=pic/pic.max() #scale to 0-1

    prediction=int(predict_class(np.array(document["image"]).reshape(1,28,28,3)))
    # display image
    plt.figure()
    plt.imshow(pic)
    plt.text(14,14,prediction,fontsize=20)
    plt.colorbar()
    plt.grid(False)
    plt.ion()
    plt.show()
    plt.pause(.7)
    plt.close()


    # read label
    print("Is this number ",prediction,"? Enter, otherwise write number: ")
    label=input()
    if label=="":
        db.gas_digit.update_one({"_id":id},{"$set":{"label":prediction}})
    else:
        label=int(label)
        if(label==10):
        # if you set label 10, delete record
            print("record deleted")
            db.gas_digit.delete_one({"_id":id})
        else:
            # update database record
            db.gas_digit.update_one({"_id":id},{"$set":{"label":label}})