app.py

import gradio as gr
import os
import cv2
from PIL import Image
import numpy as np
from matplotlib import pyplot as plt
import random
from keras.utils import get_custom_objects
import os

os.environ['SM_FRAMEWORK'] = 'tf.keras'

import segmentation_models as sm

from keras import backend as K
from keras.models import load_model

class_building = '#3C1098'
class_building = class_building.lstrip('#')
class_building = np.array(tuple(int(class_building[i:i+2], 16) for i in (0,2,4)))

class_land = '#8429F6'
class_land = class_land.lstrip('#')
class_land = np.array(tuple(int(class_land[i:i+2], 16) for i in (0,2,4)))

class_road = '#6EC1E4'
class_road = class_road.lstrip('#')
class_road = np.array(tuple(int(class_road[i:i+2], 16) for i in (0,2,4)))

class_vegetation = '#FEDD3A'
class_vegetation = class_vegetation.lstrip('#')
class_vegetation = np.array(tuple(int(class_vegetation[i:i+2], 16) for i in (0,2,4)))

class_water = '#E2A929'
class_water = class_water.lstrip('#')
class_water = np.array(tuple(int(class_water[i:i+2], 16) for i in (0,2,4)))

class_unlabeled = '#9B9B9B'
class_unlabeled = class_unlabeled.lstrip('#')
class_unlabeled = np.array(tuple(int(class_unlabeled[i:i+2], 16) for i in (0,2,4)))

def jaccard_coef(y_true, y_pred):
    y_true_flatten = K.flatten(y_true)
    y_pred_flatten = K.flatten(y_pred)
    intersection = K.sum(y_true_flatten * y_pred_flatten)
    final_coef_value = (intersection + 1.0) / (
            K.sum(y_true_flatten) + K.sum(y_pred_flatten) - intersection + 1.0)
    return final_coef_value


# six class for six weights
weights = [0.1666, 0.1666, 0.1666, 0.1666, 0.1666, 0.1666]
dice_loss = sm.losses.DiceLoss(class_weights=weights)
focal_loss = sm.losses.CategoricalFocalLoss()
total_loss = dice_loss + (1 * focal_loss)

satellite_model = load_model('satellite_segmentation_full_v2.h5',
                             custom_objects=({'dice_loss_plus_1focal_loss': total_loss, 'jaccard_coef': jaccard_coef}))


def label_to_rgb(label_segment):
    rgb_image = np.zeros((label_segment.shape[0], label_segment.shape[1], 3), dtype=np.uint8)

    rgb_image[label_segment == 0] = class_water
    rgb_image[label_segment == 1] = class_land
    rgb_image[label_segment == 2] = class_road
    rgb_image[label_segment == 3] = class_building
    rgb_image[label_segment == 4] = class_vegetation
    rgb_image[label_segment == 5] = class_unlabeled

    return rgb_image


def process_input_image(image_source):
    image = np.expand_dims(image_source, 0)
    prediction = satellite_model.predict(image)
    predicted_image = np.argmax(prediction, axis=3)
    predicted_image = predicted_image[0, :, :]

    # Convert the predicted image labels to RGB
    colored_predicted_image = label_to_rgb(predicted_image)

    return "Predicted Masked Image", colored_predicted_image


my_app = gr.Blocks()

with my_app:
    gr.Markdown("Image Processing Application UI with Gradio")
    with gr.Tabs():
        with gr.TabItem("Select your image"):
            with gr.Row():
                with gr.Column():
                    img_source = gr.Image(label="Please select source Image", shape=(256, 256))
                    source_image_loader = gr.Button("Load above Image")
                with gr.Column():
                    output_label = gr.Label(label="Image Info")
                    img_output = gr.Image(label="Image Output")
            source_image_loader.click(
                process_input_image,
                [
                    img_source
                ],
                [
                    output_label,
                    img_output
                ]
            )
my_app.launch(debug=True)