app.py

# Import libraries
import numpy as np
import matplotlib.pyplot as plt

from sklearn.datasets import fetch_olivetti_faces
from sklearn.utils.validation import check_random_state

from sklearn.ensemble import ExtraTreesRegressor
from sklearn.neighbors import KNeighborsRegressor
from sklearn.linear_model import LinearRegression
from sklearn.linear_model import RidgeCV

import gradio as gr

# Load the faces datasets
data, targets = fetch_olivetti_faces(return_X_y=True) 

train = data[targets < 30]  

n_pixels = data.shape[1] 

# Training data
# Upper half of the faces
X_train = train[:, : (n_pixels + 1) // 2] 
# Lower half of the faces
y_train = train[:, n_pixels // 2 :] 

# Fit estimators -> The problem (given half the image/features extrapolate the rest of the image/features)
ESTIMATORS = {
    "Extra trees": ExtraTreesRegressor(
        n_estimators=10, max_features=32, random_state=0
    ),
    "K-nn": KNeighborsRegressor(),
    "Linear regression": LinearRegression(),
    "Ridge": RidgeCV(),
}

for name, estimator in ESTIMATORS.items():
    estimator.fit(X_train, y_train)

test = data[targets >= 30]  
n_faces = 15
rng = check_random_state(4)
face_ids = rng.randint(test.shape[0], size=(n_faces,))
test = test[face_ids, :]

# Function for returning 64*64 image, given the image index
def imageFromIndex(index):
  return test[int(index)].reshape(1,-1).reshape(64, 64)

# Function for extrapolating face
def extrapolateFace(index, ESTIMATORS=ESTIMATORS):
  image = test[int(index)].reshape(1,-1)
  image_shape = (64, 64)
  n_cols = 1 + len(ESTIMATORS)
  n_faces = 1

  n_pixels = image.shape[1]

  # Upper half of the face
  X_upper = image[:, : (n_pixels + 1) // 2]
  # Lower half of the face
  y_ground_truth = image[:, n_pixels // 2 :]

  # y_predict: Dictionary of predicted lower-faces
  y_predict = dict() 
  for name, estimator in ESTIMATORS.items():
    y_predict[name] = estimator.predict(X_upper)
  

  plt.figure(figsize=(2.0 * n_cols, 2.5 * n_faces))
  # plt.suptitle("Face completion with multi-output estimators", size=16)


  true_face = np.hstack((X_upper, y_ground_truth))

  sub = plt.subplot(n_faces, n_cols, 1, title="true face")

  sub.axis("off")
  sub.imshow(
      true_face.reshape(image_shape), cmap=plt.cm.gray, interpolation="nearest"
  )

  for j, est in enumerate(sorted(ESTIMATORS)):
      completed_face = np.hstack((X_upper[0], y_predict[est][0]))

      sub = plt.subplot(n_faces, n_cols, 2 + j, title=est)
      sub.axis("off")
      sub.imshow(
          completed_face.reshape(image_shape),
          cmap=plt.cm.gray,
          interpolation="nearest",
    )
  
  return plt

with gr.Blocks() as demo:
    link = "https://scikit-learn.org/stable/auto_examples/miscellaneous/plot_multioutput_face_completion.html#sphx-glr-auto-examples-miscellaneous-plot-multioutput-face-completion-py"
    title = "Face completion with a multi-output estimators"
    gr.Markdown(f"# {title}")
    gr.Markdown(f"### This demo is based on this [scikit-learn example]({link}).")
    gr.Markdown("### In this demo, we compare 4 multi-output estimators to complete images. \
    The goal is to predict the lower half of a face given its upper half.")

    gr.Markdown("#### Use the below slider to choose a face's image. \
    Consequently, observe how the four estimators complete the lower half of that face.")

    with gr.Row():
        with gr.Column(scale=1):  
          image_index = gr.Slider(1,15,1,step=1, label="Image Index", info="Choose an image") 
          face_image = gr.Image()   
        with gr.Column(scale=2):
          plot = gr.Plot(label=f"Face completion with multi-output estimators")

    image_index.change(imageFromIndex, inputs=[image_index], outputs=[face_image])
    image_index.change(extrapolateFace, inputs=[image_index], outputs=[plot])
    demo.load(imageFromIndex, inputs=[image_index], outputs=[face_image])
    demo.load(extrapolateFace, inputs=[image_index], outputs=[plot])

if __name__ == "__main__":
    demo.launch(debug=True)