app.py

import streamlit as st

# x = st.slider("Select a value")
# st.write(x, "squared is", x * x)
st.title('Welcome to the Labelbox custom classifier training application!')
st.header('In this module you will be able to add data to your instance if you dont already have it, and then use bulk classification to train a custom image classification model')
st.subheader("If you don't have data in your org, enter your API Click the button below! Otherwise, Skip to section 2")
# -*- coding: utf-8 -*-
"""
Original file is located at
    https://colab.research.google.com/drive/1nOSff67KXhNgX_XSfnv3xnddobRoaK0d

"""

api_key = st.text_input("Enter your api key:", type="password")

import labelbox
import labelpandas as lp
import os
import pandas as pd
from tensorflow.python.lib.io import file_io
import io
from pandas import read_csv

# read csv file from google cloud storage
def read_data(gcs_path):
   file_stream = file_io.FileIO(gcs_path, mode='r')
   csv_data = read_csv(io.StringIO(file_stream.read()))
   return csv_data
def freedatatolb(amount_of_data):
    client = lp.Client(api_key)
    gcs_path = 'https://storage.googleapis.com/solution_accelerator_datasets/images_styles.csv'
    df = pd.read_csv(gcs_path)
    df = df.drop(['id', 'season', 'usage', 'year',"gender", "masterCategory", "subCategory", "articleType","baseColour"], axis =1)
    fields ={"row_data":["link"], # Column containing URL to asset (single)
        "global_key": ['filename'], # Column containing globalkey value (single, unique)
        "external_id": ["productDisplayName"], # Column containing external ID value (single)
        "metadata_string": [], # Column containing string metadata values (multiple)
        "metadata_number": [], # Column containing number metadata values (multiple)
        "metadata_datetime": [] # Column containing datetime metadata values (multiple, must be ISO 8601)
        }
    columns = {}

    for field in fields.keys():
        for name in fields[field]:
            if field.startswith('metadata'):
                columns[name] = f"{field.split('_')[0]}///{field.split('_')[1]}///{name}"
            else:
                columns[name] = field
    new_df = df.rename(columns=(columns))
    testdf = new_df.head(amount_of_data)
    dataset_id = client.lb_client.create_dataset(name = str(gcs_path.split('/')[-1])).uid
# dataset_id = client.lb_client.get_dataset("c4b7prd6207850000lljx2hr8").uid
    results = client.create_data_rows_from_table(
        table = testdf,
        dataset_id = dataset_id,
        skip_duplicates = True, # If True, will skip data rows where a global key is already in use,
        verbose = True, # If True, prints information about code execution
    )
    return results
data_amount = st.slider("choose amout of data to add to labelbox", 250, 1000)
if st.button("Add data to your Labelbox"):
    st.write(f"adding {data_amount} datarows to Labelbox instance")
    bing = freedatatolb(data_amount)
    st.write(bing)
             
st.title("SECTION 2")
st.title("Auto Image classifier training and inference: Imagnet Weights")

# -*- coding: utf-8 -*-
"""
Original file is located at
    https://colab.research.google.com/drive/1CSyAE9DhwGTl7bLaSoo7QSyMuoEqJpCj

"""

def train_and_inference(api_key, ontology_id, model_run_id):
    # st.write('thisisstarting')
    api_key = api_key # insert Labelbox API key
    ontology_id = ontology_id # get the ontology ID from the Settings tab at the top left of your model run
    model_run_id = model_run_id #get the model run ID from the settings gear icon on the right side of your Model Run
    # st.write('1')
    import pydantic
    st.write(pydantic.__version__)

    import numpy as np
    # st.write('2')
    import tensorflow as tf
    # st.write('3')
    from tensorflow.keras import layers
    # st.write('4')
    from tensorflow.keras.models import Sequential
    # st.write('5')
    from tensorflow.keras.preprocessing.image import ImageDataGenerator
    # st.write('6')
    import os
    # st.write('7')
    import labelbox
    # st.write('zat')
    from labelbox import Client
    # st.write('8')
    # st.write('9')
    import numpy as np
    import tensorflow as tf
    from tensorflow.keras import layers
    from tensorflow.keras.models import Sequential
    from tensorflow.keras.preprocessing.image import ImageDataGenerator
    import os
    from labelbox.schema.ontology import OntologyBuilder, Tool, Classification, Option
    from labelbox import Client, LabelingFrontend, LabelImport, MALPredictionImport
    from labelbox.data.annotation_types import (
        Label, ImageData, ObjectAnnotation, MaskData,
        Rectangle, Point, Line, Mask, Polygon,
        Radio, Checklist, Text,
        ClassificationAnnotation, ClassificationAnswer
    )
    from labelbox import MediaType
    from labelbox.data.serialization import NDJsonConverter
    import pandas as pd
    import shutil
    import labelbox.data
    import scipy

    import json
    import uuid
    import time
    import requests
    import pandas as pd
    import shutil

    import json
    import uuid
    import time
    import requests
    # st.write('imports')

    """Connect to labelbox client
    Define Model Variables
    """

    client = Client(api_key)
    EPOCHS = 10

    """#Setup Training

    Export Classifications from Model Run
    """

    model_run = client.get_model_run(model_run_id)

    client.enable_experimental = True
    data_json = model_run.export_labels(download=True)
    print(data_json)

    """Separate datarows into folders."""

    import requests
    import os

    from urllib.parse import unquote

    def download_and_save_image(url, destination_folder, filename):
        try:
            # Decode the URL
            url = unquote(url)

            # Ensure destination directory exists
            if not os.path.exists(destination_folder):
                os.makedirs(destination_folder)

            # Start the download process
            response = requests.get(url, stream=True)

            # Check if the request was successful
            if response.status_code == 200:
                file_path = os.path.join(destination_folder, filename)
                with open(file_path, 'wb') as file:
                    for chunk in response.iter_content(8192):
                        file.write(chunk)
            #     st.write(f"Image downloaded and saved: {file_path}")
            # else:
            #     st.write(f"Failed to download the image. Status code: {response.status_code}")
        except Exception as e:
            st.write(f"An error occurred: {e}")



    BASE_DIR = 'dataset'
    labeldict = {}

    for entry in data_json:
        data_split = entry['Data Split']
        if data_split not in ['training', 'validation']:  # we are skipping 'test' for now
            continue

        image_url = f"{entry['Labeled Data']}"
        label = entry['Label']['classifications'][0]['answer']['value']
        labeldict[label] = entry['Label']['classifications'][0]['answer']['title']

        destination_folder = os.path.join(BASE_DIR, data_split, label)
        filename = os.path.basename(image_url)
        # st.write(filename)

        download_and_save_image(image_url, destination_folder, filename)

    """#Train Model"""
    st.write(labeldict)

    import tensorflow as tf
    from tensorflow.keras.preprocessing.image import ImageDataGenerator
    from tensorflow.keras.applications import MobileNetV2
    from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
    from tensorflow.keras.models import Model
    from tensorflow.keras.optimizers import Adam

    TRAIN_DIR = 'dataset/training'
    VALIDATION_DIR = 'dataset/validation'
    IMG_HEIGHT, IMG_WIDTH = 224, 224  # default size for MobileNetV2
    BATCH_SIZE = 32

    train_datagen = ImageDataGenerator(
        rescale=1./255,
        rotation_range=20,
        width_shift_range=0.2,
        height_shift_range=0.2,
        shear_range=0.2,
        zoom_range=0.2,
        horizontal_flip=True,
        fill_mode='nearest'
    )

    validation_datagen = ImageDataGenerator(rescale=1./255)

    train_ds = train_datagen.flow_from_directory(
        TRAIN_DIR,
        target_size=(IMG_HEIGHT, IMG_WIDTH),
        batch_size=BATCH_SIZE,
        class_mode='categorical'
    )

    validation_ds = validation_datagen.flow_from_directory(
        VALIDATION_DIR,
        target_size=(IMG_HEIGHT, IMG_WIDTH),
        batch_size=BATCH_SIZE,
        class_mode='categorical'
    )

    base_model = MobileNetV2(input_shape=(IMG_HEIGHT, IMG_WIDTH, 3),
                            include_top=False,
                            weights='imagenet')

    # Freeze the base model
    for layer in base_model.layers:
        layer.trainable = False

    # Create custom classification head
    x = base_model.output
    x = GlobalAveragePooling2D()(x)
    x = Dense(1024, activation='relu')(x)
    predictions = Dense(train_ds.num_classes, activation='softmax')(x)

    model = Model(inputs=base_model.input, outputs=predictions)

    model.compile(optimizer=Adam(learning_rate=0.0001),
                loss='categorical_crossentropy',
                metrics=['accuracy'])

    st.write("training")
    history = model.fit(
        train_ds,
        validation_data=validation_ds,
        epochs=EPOCHS
    )

    """Run Inference on Model run Datarows"""
    st.write('running Inference')

    import numpy as np
    import requests
    from tensorflow.keras.preprocessing import image
    from PIL import Image
    from io import BytesIO
    # Fetch the image from the URL
    def load_image_from_url(img_url, target_size=(224, 224)):
        response = requests.get(img_url)
        img = Image.open(BytesIO(response.content))
        img = img.resize(target_size)
        img_array = image.img_to_array(img)
        return np.expand_dims(img_array, axis=0)
    def make_prediction(img_url):
    # Image URL
        img_url = img_url

        # Load and preprocess the image
        img_data = load_image_from_url(img_url)
        img_data = img_data / 255.0  # Normalize the image data to [0,1]

        # Make predictions
        predictions = model.predict(img_data)
        predicted_class = np.argmax(predictions[0])

        # Retrieve the confidence score (probability) for the predicted class
        confidence = predictions[0][predicted_class]

        # Map the predicted class index to its corresponding label
        class_map = train_ds.class_indices
        inverse_map = {v: k for k, v in class_map.items()}
        predicted_label = inverse_map[predicted_class]

        return predicted_label, confidence

    from tensorflow.errors import InvalidArgumentError  # Add this import
    ontology = client.get_ontology(ontology_id)
    label_list = []
    st.write(ontology)
    for datarow in model_run.export_labels(download=True):
        try:
            label, confidence = make_prediction(datarow['Labeled Data'])
        except InvalidArgumentError as e:
            print(f"InvalidArgumentError: {e}. Skipping this data row.")
            continue  # Skip to the next datarow if an exception occurs
        my_checklist_answer = ClassificationAnswer(
                    name = labeldict[label.lower()],
                    confidence=confidence)
        checklist_prediction = ClassificationAnnotation(
    name=ontology.classifications()[0].instructions,
        value=Radio(
            answer = my_checklist_answer
    ))
    # print(datarow["DataRow ID"])
        label_prediction = Label(
        data=ImageData(uid=datarow['DataRow ID']),
        annotations = [checklist_prediction])
        label_list.append(label_prediction)

    prediction_import = model_run.add_predictions(
        name="prediction_upload_job"+str(uuid.uuid4()),
        predictions=label_list)

    prediction_import.wait_until_done()

    st.write(prediction_import.errors == [])
    if prediction_import.errors == []:
        return "Model Trained and inference ran successfully"
    else:
        return prediction_import.errors

st.title("Enter Applicable IDs and keys below")

model_run_id = st.text_input("Enter your model run ID:")
ontology_id = st.text_input("Enter your ontology ID:")

if st.button("Train and run inference"):
    st.write('Starting Up...')
        # Check if the key is not empty
    if api_key + model_run_id + ontology_id:
        result = train_and_inference(api_key, ontology_id, model_run_id)
        st.write(result)
    else:
        st.warning("Please enter all keys.")