app.py

import numpy as np
import streamlit as st
from PIL import Image
from streamlit_image_select import image_select
import base64
import cv2
import tensorflow as tf

# set title of page
st.markdown("<h1 '><font color='red' ><center>Emotion Detector App</center></h1>",unsafe_allow_html=True)


# ctreating gride of image to select
img = image_select(
    label="_Select a image to predict_",
    images=[
        "images/1.jpg",
        "images/2.jpeg",
        "images/3.jpeg",
        "images/4.jpg"

    ],

)

# image select for background
titleimg = "giphy (1).gif"

#impliment background formating
def set_bg_hack(main_bg):
    # set bg name
    main_bg_ext = "gif"
    st.markdown(
        f"""
         <style>
         .stApp {{
             background: url(data:image/{main_bg_ext};base64,{base64.b64encode(open(main_bg, "rb").read()).decode()});
             background-repeat: no-repeat;
             background-position: right 50% bottom 95% ;
             background-size: cover;
             background-attachment: scroll;
         }}
         </style>
         """,
        unsafe_allow_html=True,
    )

set_bg_hack(titleimg)

uploaded_img = st.file_uploader("Upload an image file",
                                type  = ["png", "jpg", "jpeg"])


# load model
@st.cache_resource
def cache_model(model_add):
    model = tf.keras.models.load_model(model_add)
    return model


model = cache_model("emotion_detector")

# creating predict button
predict = st.button("Predict")

# defining harcascade classifier and class_names
face_detector = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
class_names = ["Angry", "Happy", "Sad"]


def model_pred(model, image):
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    results = face_detector.detectMultiScale(gray, scaleFactor=1.05,
                                             minNeighbors=10,
                                             minSize=(100, 100))
    if len(results) != 0:
        for x, y, w, h in results:
            img_crp = image[y:y + h, x:x + w]
            img_crp = cv2.resize(img_crp, (350, 350))
            y_pred_prob = model.predict(tf.expand_dims(img_crp,
                                                       axis=0))
            y_pred = np.argmax(y_pred_prob, axis=-1)
            # print(y_pred_prob)
            label = class_names[int(y_pred)]
            cv2.rectangle(image, (x, y), (x + w, y + h),
                          color=(0, 255, 0),
                          thickness=10)
            cv2.putText(image, f"{label},{round(round(np.max(y_pred_prob), 2)*100,2)}%",
                        (x, y + h), cv2.FONT_HERSHEY_COMPLEX, 2,
                        (0, 255, 255), 2)
    return image


if predict:
    if uploaded_img:
        # img_array = np.array(uploaded_img)
        img_array = np.array(Image.open(uploaded_img))
        result_img = model_pred(model, img_array)
        st.image(result_img)

        # image select for background
        titleimg = "CLOUD.png"
        
        #impliment background formating
        def set_bg_hack(main_bg):
            # set bg name
            main_bg_ext = "png"
            st.markdown(
                f"""
                 <style>
                 .stApp {{
                     background: url(data:image/{main_bg_ext};base64,{base64.b64encode(open(main_bg, "rb").read()).decode()});
                     background-repeat: no-repeat;
                     background-position: right 50% bottom 95% ;
                     background-size: cover;
                     background-attachment: scroll;
                 }}
                 </style>
                 """,
                unsafe_allow_html=True,
            )
        
        set_bg_hack(titleimg)

    else:
        st.write("Please upload a valid image")

else:
    image_array = np.array(Image.open(img))
    result_img = model_pred(model, image_array)
    st.image(result_img)