app.py

import os
import random
import numpy as np
import tensorflow as tf
from PIL import Image

import gradio as gr
from huggingface_hub import from_pretrained_keras

model = from_pretrained_keras("keras-io/GauGAN-Image-generation")
  
def predict(image_file, segmentation_png, bitmap_img):

  image_list = [segmentation_png, image_file, bitmap_img]

  image = tf.image.decode_png(tf.io.read_file(image_list[1]), channels=3)
  image = tf.cast(image, tf.float32) / 127.5 - 1

  segmentation_file = tf.image.decode_png(tf.io.read_file(image_list[0]), channels=3)
  segmentation_file = tf.cast(segmentation_file, tf.float32)/127.5 - 1

  label_file = tf.image.decode_bmp(tf.io.read_file(image_list[2]), channels=0)
  # label_file = tf.image.rgb_to_grayscale(tf.image.decode_bmp(tf.io.read_file(image_list[2]), channels=3))
  # print("after decode_bmp --> ", label_file.shape, type(label_file))
  label_file = tf.squeeze(label_file)

  image_list = [segmentation_file, image, label_file]

  crop_size = tf.convert_to_tensor((256, 256))

  image_shape = tf.shape(image_list[1])[:2]

  margins = image_shape - crop_size

  y1 = tf.random.uniform(shape=(), maxval=margins[0], dtype=tf.int32)
  x1 = tf.random.uniform(shape=(), maxval=margins[1], dtype=tf.int32)
  y2 = y1 + crop_size[0]
  x2 = x1 + crop_size[1]

  cropped_images = []
  for img in image_list:
    cropped_images.append(img[y1:y2, x1:x2])

  final_img_list = [tf.expand_dims(cropped_images[0], axis=0), tf.expand_dims(cropped_images[1], axis=0), tf.expand_dims(tf.one_hot(cropped_images[2], 12), axis=0)]

  # print(final_img_list[0].shape)
  # print(final_img_list[1].shape)
  # print(final_img_list[2].shape)

  latent_vector = tf.random.normal(shape=(1, 256), mean=0.0, stddev=2.0)

  # Generate fake images
  fake_image = model.predict([latent_vector, final_img_list[2]])
  fake_img = tf.squeeze(fake_image, axis=0)

  return np.array((fake_img+1)/2)
  
# input
input = [gr.inputs.Image(type="filepath", label="Ground Truth - Real Image (jpg)"), 
         gr.inputs.Image(type="filepath", label="Corresponding Segmentation (png)"),
         gr.inputs.Image(type="filepath", label="Corresponding bitmap image (bmp)", image_mode="L")]

examples = [["facades_data/cmp_b0010.jpg", "facades_data/cmp_b0010.png", "facades_data/cmp_b0010.bmp"],
            ["facades_data/cmp_b0020.jpg", "facades_data/cmp_b0020.png", "facades_data/cmp_b0020.bmp"],
            ["facades_data/cmp_b0030.jpg", "facades_data/cmp_b0030.png", "facades_data/cmp_b0030.bmp"],
            ["facades_data/cmp_b0040.jpg", "facades_data/cmp_b0040.png", "facades_data/cmp_b0040.bmp"], 
            ["facades_data/cmp_b0050.jpg", "facades_data/cmp_b0050.png", "facades_data/cmp_b0050.bmp"]]

# output
output = [gr.outputs.Image(type="numpy", label="Generated - Conditioned Images")]

title = "GauGAN For Conditional Image Generation"
description = "Upload an Image or take one from examples to generate realistic images that are conditioned on cue images and segmentation maps"

gr.Interface(fn=predict, inputs = input, outputs = output, examples=examples, allow_flagging=False, analytics_enabled=False,
  title=title, description=description, article="<center>Space By: <u><a href='https://github.com/robotjellyzone'><b>Kavya Bisht</b></a></u> \n Based on <a href='https://keras.io/examples/generative/gaugan/'><b>this notebook</b></a></center>").launch(enable_queue=True, debug=True)