app.py

import gradio as gr
import numpy as np
from PIL import Image
import tensorflow as tf


def iou(y_true, y_pred):
    def f(y_true, y_pred):
        intersection = (y_true * y_pred).sum()
        union = y_true.sum() + y_pred.sum() - intersection
        x = (intersection + 1e-15) / (union + 1e-15)
        x = x.astype(np.float32)
        return x
    return tf.numpy_function(f, [y_true, y_pred], tf.float32)

def dice_coef(y_true, y_pred):
    y_true = tf.keras.layers.Flatten()(y_true)
    y_pred = tf.keras.layers.Flatten()(y_pred)
    intersection = tf.reduce_sum(y_true * y_pred)
    return (2. * intersection + 1e-15) / (tf.reduce_sum(y_true) + tf.reduce_sum(y_pred))

def dice_loss(y_true, y_pred):
    return 1.0 - dice_coef(y_true, y_pred)

def read_image(file_path, target_size=(256, 256)):
    img = Image.open(file_path)
    img = img.resize(target_size)
    x = np.array(img, dtype=np.float32)
    x = x / 255.0
    return x

def preprocess_image(img):
    if img.shape[-1] == 4:
        img = img[..., :3]
    img_expanded = np.expand_dims(img, axis=0)
    return img_expanded

def predict_image(model, img):
    pred = model.predict(img)
    return pred[0, ...]  # Taking the first item in the batch

# Load the model with specific custom objects
loaded_model = tf.keras.models.load_model(
    "oryx_road_segmentation_model.h5",
    custom_objects={'dice_coef': dice_coef, 'iou': iou})

def process_image(image):
    img = read_image(image)
    img_preprocessed = preprocess_image(img)
    pred = predict_image(loaded_model, img_preprocessed)

    # Convert single-channel image to RGB by duplicating the channel across RGB
    pred_img = np.squeeze(pred)  # Remove the singleton dimension
    pred_img = np.clip(pred_img, 0, 1)  # Ensure all values are between 0 and 1
    pred_img_rgb = np.stack((pred_img,)*3, axis=-1)  # Stack grayscale across three channels to mimic RGB
    pred_img_rgb = (pred_img_rgb * 255).astype(np.uint8)  # Scale to 0-255 and convert to uint8
    return Image.fromarray(pred_img_rgb)  # Now converting a proper 2D RGB array

# Sample images directory or paths
sample_images = ["234989_sat.jpg", "751359_sat.jpg","168243_sat.jpg","877873_sat.jpg","836987_sat.jpg"]

# Gradio Interface
iface = gr.Interface(
    fn=process_image,
    inputs=gr.Image(type="filepath"),
    outputs=gr.Image(type="pil", label="Predicted Image"),
    title="orYx Models' - Road Segmentation Predictor",
    description="Upload an image or choose a sample and view the model's segmentation for roads on different terrains.",
    examples= sample_images
)

iface.launch(debug=True)