app.py

import os, io
import cv2
import gradio as gr
import tensorflow as tf
import numpy as np
import keras.backend as K

from matplotlib import pyplot as plt
from PIL import Image
from tensorflow import keras


resized_shape = (768, 768, 3)
IMG_SCALING = (1, 1)


# # Download the model file
# def download_model():
#     url = "https://drive.google.com/uc?id=1FhICkeGn6GcNXWTDn1s83ctC-6Mo1UXk"
#     output = "seg_unet_model.h5"
#     gdown.download(url, output, quiet=False)
#     return output

model_file = "./seg_unet_model.h5"

#Custom objects for model

def Combo_loss(y_true, y_pred, eps=1e-9, smooth=1):
    targets = tf.dtypes.cast(K.flatten(y_true), tf.float32)
    inputs = tf.dtypes.cast(K.flatten(y_pred), tf.float32)
    intersection = K.sum(targets * inputs)
    dice = (2. * intersection + smooth) / (K.sum(targets) + K.sum(inputs) + smooth)
    inputs = K.clip(inputs, eps, 1.0 - eps)
    out = - (ALPHA * ((targets * K.log(inputs)) + ((1 - ALPHA) * (1.0 - targets) * K.log(1.0 - inputs))))
    weighted_ce = K.mean(out, axis=-1)
    combo = (CE_RATIO * weighted_ce) - ((1 - CE_RATIO) * dice)
    return combo

def dice_coef(y_true, y_pred, smooth=1):
    y_pred = tf.dtypes.cast(y_pred, tf.int32)
    y_true = tf.dtypes.cast(y_true, tf.int32)
    intersection = K.sum(y_true * y_pred, axis=[1,2,3])                     
    union = K.sum(y_true, axis=[1,2,3]) + K.sum(y_pred, axis=[1,2,3])           
    return K.mean((2 * intersection + smooth) / (union + smooth), axis=0)

def focal_loss_fixed(y_true, y_pred, gamma=2.0, alpha=0.25):
    pt_1 = tf.where(tf.equal(y_true, 1), y_pred, tf.ones_like(y_pred))
    pt_0 = tf.where(tf.equal(y_true, 0), y_pred, tf.zeros_like(y_pred))
    focal_loss_fixed = -K.mean(alpha * K.pow(1. - pt_1, gamma) * K.log(pt_1+K.epsilon())) - K.mean((1 - alpha) * K.pow(pt_0, gamma) * K.log(1. - pt_0 + K.epsilon()))
    return focal_loss_fixed

# Load the model
seg_model = keras.models.load_model('seg_unet_model.h5', custom_objects={'Combo_loss': Combo_loss, 'focal_loss_fixed': focal_loss_fixed, 'dice_coef': dice_coef})

# inputs = gr.inputs.Image(type="pil", label="Upload an image")
# image_output = gr.outputs.Image(type="pil", label="Output Image")
# outputs = gr.outputs.HTML() #uncomment for single class output 

rows = 1
columns = 1

def gen_pred(img, model=seg_model):
    pil_image = img.convert('RGB')
    open_cv_image = np.array(pil_image)
    img = open_cv_image[:, :, ::-1].copy() 
    # img = cv2.imread("./003e2c95d.jpg")
    img = img[::IMG_SCALING[0], ::IMG_SCALING[1]]
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    img = img/255
    img = tf.expand_dims(img, axis=0)
    pred = model.predict(img)
    pred = np.squeeze(pred, axis=0)
    fig = plt.figure(figsize=(3, 3))
    fig.add_subplot(rows, columns, 1)
    # plt.imshow(pred, interpolation='catrom')
    plt.imshow(pred)
    plt.axis('off')
    plt.show()
    return fig

title = "<h1 style='text-align: center;'>Semantic Segmentation (Airbus Ship Detection Challenge)</h1>"
description = "Upload an image and get prediction mask"

gr.Interface(fn=gen_pred, 
             inputs=[gr.Image(type='pil')], 
             outputs=["plot"], 
             title=title, 
             examples=[["00c3db267.jpg"], ["00dc34840.jpg"], ["00371aa92.jpg"]],
             description=description,
            enable_queue=True).launch()