AnJ47/CNN_Caption_generator

1import streamlit as st2from tensorflow import keras3from keras.models import load_model4from keras.preprocessing.text import Tokenizer5from keras.utils import pad_sequences6from keras.applications.vgg16 import VGG16, preprocess_input7from keras.models import Model8from keras.layers import Input, Dense, LSTM ,add,Dropout, Embedding 9import numpy as np10import cv211from PIL import Image, ImageOps12 13 14# Obtained earlier 15# MAX_LENGTH=3516    17def idx_to_word(integer,tokenizer):18    for word, index in tokenizer.word_index.items():19        if index==integer:20            return word21    return None22 23def read_list(file_path):24    lst = []25    with open(file_path, 'r') as file:26        for line in file:27            lst.append(line.strip())28    return lst29 30def preprocess_image(img):31    size = (224,224)  32    image = ImageOps.fit(img, size, Image.ANTIALIAS)33    image = np.asarray(image)34    img = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)35    img_reshape = img[np.newaxis,...]36    image = preprocess_input(img_reshape)37    return image38 39def predict_caption(model,feature,tokenizer):40    max_length =3541    # add start tag for generation process 42    in_text='startseq'43    #iterate over the max length of sequence44    for i in range(max_length):45        # Encode input sequence46        sequence=tokenizer.texts_to_sequences([in_text])[0]47        # Pad the sequence48        sequence= pad_sequences([sequence],max_length)49        # Predict the next word50        # image=image.reshape((1,4096))51        yhat = model.predict([feature,sequence],verbose=0) 52        # Get the index with the highest probability53        yhat=np.argmax(yhat)54        # Convert the index to word55        word = idx_to_word(yhat,tokenizer)56        if word is None:57            break58        # Append word as input59        in_text+= " "+word60        # Stop if we reach end tag61        if word=='endseq':62            break63    return in_text64 65def generate_clean_caption(in_text):66    # Split the string into a list of words67    words = in_text.split()68    # Join all words except the first and last69    result = ' '.join(words[1:-1])  70    return result71    72    73 74## Layout ##75rad = st.sidebar.radio("Navigation",["Home","Caption Generator"])76 77 78if rad=="Home":79    st.title("CNN Caption Generator")80    st.subheader("By Anuraj Bhaskar")81    st.markdown("""Upload an Image in the Caption Generator page to generate Caption""")82    83    # st.image("sample.jpg")84    85elif rad=="Caption Generator":86    st.write("""87            # Caption Generation88    """)89   90    vocab_size =848591    vgg = VGG16()92    # restructure the model93    vgg_model = Model(inputs=vgg.inputs, outputs=vgg.layers[-2].output)94  95    with st.spinner("Loading Model..."):96        model = load_model("model.h5") 97     98    file = st.file_uploader("Upload an Image")99    st.write(file)100    st.set_option('deprecation.showfileUploaderEncoding', False)101        102    if file != None:103        image = Image.open(file)104        st.image(image,caption="Uploaded Image",use_column_width=True)105        #  Prepocessing the Image106        image=preprocess_image(image)107        feature =vgg_model.predict(image,verbose=0)108        109        all_captions = read_list('all_captions.txt')110        #Tokenize the text111        tokenizer=Tokenizer()112        tokenizer.fit_on_texts(all_captions)113        114        115       116        # feature = vgg_model.predict(image,verbose=0)117        118        in_text=predict_caption(model,feature,tokenizer) 119        120        # Removing Startseq and Endseq121        caption= generate_clean_caption(in_text)122        123        # # Display the Caption124        st.markdown('<b>Generated Caption:</b>',True)125        st.markdown(caption.upper())126    else:127        st.write("No file uploaded")128