File size: 3,512 Bytes
72a1628
 
 
 
 
 
 
 
 
e4fb230
79c6687
 
 
 
 
 
 
 
 
e47598c
72a1628
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e4fb230
72a1628
e4fb230
72a1628
e4fb230
79c6687
 
 
 
 
 
 
 
 
 
 
e4fb230
72a1628
 
 
 
 
 
 
 
e4fb230
72a1628
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
# Import libraries

import cv2
from tensorflow import keras
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
import segmentation_models as sm

def get_mask(image:Image) -> Image:
    """
    This function generates a mask of the image that highlights all the sofas in the image.
    This uses a pre-trained Unet model with a resnet50 backbone.
    Remark: The model was trained on 640by640 images and it is therefore best that the image has the same size.
    Parameters:
            image = original image
    Return:
            mask  = corresponding maks of the image
    """
    model_path = "model_checkpoint.h5"
    CLASSES = ['sofa']
    BACKBONE = 'resnet50'

    # define network parameters
    n_classes = 1 if len(CLASSES) == 1 else (len(CLASSES) + 1)  # case for binary and multiclass segmentation
    activation = 'sigmoid' if n_classes == 1 else 'softmax'
    preprocess_input = sm.get_preprocessing(BACKBONE)
    sm.set_framework('tf.keras')
    LR=0.0001

    #create model architecture
    model = sm.Unet(BACKBONE, classes=n_classes, activation=activation)
    # define optomizer
    optim = keras.optimizers.Adam(LR)
    # Segmentation models losses can be combined together by '+' and scaled by integer or float factor
    dice_loss = sm.losses.DiceLoss()
    focal_loss = sm.losses.BinaryFocalLoss() if n_classes == 1 else sm.losses.CategoricalFocalLoss()
    total_loss = dice_loss + (1 * focal_loss)
    # actulally total_loss can be imported directly from library, above example just show you how to manipulate with losses
    # total_loss = sm.losses.binary_focal_dice_loss # or sm.losses.categorical_focal_dice_loss 
    metrics = [sm.metrics.IOUScore(threshold=0.5), sm.metrics.FScore(threshold=0.5)]
    # compile keras model with defined optimozer, loss and metrics
    model.compile(optim, total_loss, metrics)

    #load model
    model.load_weights(model_path) 

    test_img = np.array(image)#cv2.imread(path, cv2.IMREAD_COLOR)       
    test_img = cv2.resize(test_img, (640, 640))
    test_img = cv2.cvtColor(test_img, cv2.COLOR_RGB2BGR)
    test_img = np.expand_dims(test_img, axis=0)

    prediction = model.predict(preprocess_input(np.array(test_img))).round()
    mask = Image.fromarray(prediction[...,0].squeeze()*255).convert("L")
    return mask 

def replace_sofa(image:Image, mask:Image, styled_sofa:Image) -> Image:
    """
    This function replaces the original sofa in the image by the new styled sofa according
    to the mask.
    Remark: All images should have the same size.
    Input:
        image       = Original image
        mask        = Generated masks highlighting the sofas in the image
        styled_sofa = Styled image
    Return:
        new_image   = Image containing the styled sofa
    """
    image,mask,styled_sofa = np.array(image),np.array(mask),np.array(styled_sofa) 
    #image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    styled_sofa = cv2.cvtColor(styled_sofa, cv2.COLOR_BGR2RGB)

    _, mask = cv2.threshold(mask, 10, 255, cv2.THRESH_BINARY)
    mask_inv = cv2.bitwise_not(mask)
    image_bg = cv2.bitwise_and(image,image,mask = mask_inv)
    sofa_fg = cv2.bitwise_and(styled_sofa,styled_sofa,mask = mask)
    new_image = cv2.add(image_bg,sofa_fg)
    return Image.fromarray(new_image)

# image = cv2.imread('input/sofa.jpg')
# mask = cv2.imread('masks/sofa.jpg')
# styled_sofa = cv2.imread('output/sofa_stylized_style.jpg')

# #get_mask(image)

# plt.imshow(replace_sofa(image,mask,styled_sofa))
# plt.show()