File size: 6,302 Bytes
d11a7a2
d360108
 
 
62248f2
 
d11a7a2
62248f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d11a7a2
d360108
 
3858779
d360108
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c80976f
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
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
import gradio as gr
import torch
from torchvision.transforms import transforms
import numpy as np
from typing import Optional
import torch.nn as nn

class BasicBlock(nn.Module):
    """ResNet Basic Block.

    Parameters
    ----------
    in_channels : int
        Number of input channels
    out_channels : int
        Number of output channels
    stride : int, optional
        Convolution stride size, by default 1
    identity_downsample : Optional[torch.nn.Module], optional
        Downsampling layer, by default None
    """

    def __init__(self,
                in_channels: int,
                out_channels: int,
                stride: int = 1,
                identity_downsample: Optional[torch.nn.Module] = None):
        super(BasicBlock, self).__init__()
        self.conv1 = nn.Conv2d(in_channels,
                              out_channels,
                              kernel_size = 3,
                              stride = stride,
                              padding = 1)
        self.bn1 = nn.BatchNorm2d(out_channels)
        self.relu = nn.ReLU()
        self.conv2 = nn.Conv2d(out_channels,
                              out_channels,
                              kernel_size = 3,
                              stride = 1,
                              padding = 1)
        self.bn2 = nn.BatchNorm2d(out_channels)
        self.identity_downsample = identity_downsample

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """Apply forward computation."""
        identity = x
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.conv2(x)
        x = self.bn2(x)

        # Apply an operation to the identity output.
        # Useful to reduce the layer size and match from conv2 output
        if self.identity_downsample is not None:
            identity = self.identity_downsample(identity)
        x += identity
        x = self.relu(x)
        return x

class ResNet18(nn.Module):
    """Construct ResNet-18 Model.

    Parameters
    ----------
    input_channels : int
        Number of input channels
    num_classes : int
        Number of class outputs
    """

    def __init__(self, input_channels, num_classes):

        super(ResNet18, self).__init__()
        self.conv1 = nn.Conv2d(input_channels,
                               64, kernel_size = 7,
                              stride = 2, padding=3)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU()
        self.maxpool = nn.MaxPool2d(kernel_size = 3,
                                   stride = 2,
                                   padding = 1)

        self.layer1 = self._make_layer(64, 64, stride = 1)
        self.layer2 = self._make_layer(64, 128, stride = 2)
        self.layer3 = self._make_layer(128, 256, stride = 2)
        self.layer4 = self._make_layer(256, 512, stride = 2)

        # Last layers
        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(512, num_classes)

    def identity_downsample(self, in_channels: int, out_channels: int) -> nn.Module:
        """Downsampling block to reduce the feature sizes."""
        return nn.Sequential(
             nn.Conv2d(in_channels,
                       out_channels,
                       kernel_size = 3,
                       stride = 2,
                       padding = 1),
            nn.BatchNorm2d(out_channels)
        )

    def _make_layer(self, in_channels: int, out_channels: int, stride: int) -> nn.Module:
        """Create sequential basic block."""
        identity_downsample = None

        # Add downsampling function
        if stride != 1:
            identity_downsample = self.identity_downsample(in_channels, out_channels)

        return nn.Sequential(
                    BasicBlock(in_channels, out_channels, identity_downsample=identity_downsample, stride=stride),
                    BasicBlock(out_channels, out_channels)
                    )

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        x = self.avgpool(x)
        x = x.view(x.shape[0], -1)
        x = self.fc(x)
        return x

model = ResNet18(1, 5)

checkpoint = torch.load('acc=0.94.ckpt', map_location=torch.device('cpu'))

# The state dict will contains net.layer_name
# Our model doesn't contains `net.` so we have to rename it
state_dict = checkpoint['state_dict']
for key in list(state_dict.keys()):
    if 'net.' in key:
        state_dict[key.replace('net.', '')] = state_dict[key]
        del state_dict[key]

model.load_state_dict(state_dict)
model.eval()

class_names = ['abdominal', 'adult', 'others', 'pediatric', 'spine']
class_names.sort()

transformation_pipeline = transforms.Compose([
    transforms.ToPILImage(),
    transforms.Grayscale(num_output_channels=1),
    transforms.CenterCrop((384, 384)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.50807575], std=[0.20823])
])


def preprocess_image(image: np.ndarray):
    """Preprocess the input image.

    Note that the input image is in RGB mode.

    Parameters
    ----------
    image: np.ndarray
        Input image from callback.
    """

    image = transformation_pipeline(image)
    image = torch.unsqueeze(image, 0)

    return image


def image_classifier(inp):
    """Image Classifier Function.

    Parameters
    ----------
    inp: Optional[np.ndarray] = None
        Input image from callback

    Returns
    -------
    Dict
        A dictionary class names and its probability
    """

    # If input not valid, return dummy data or raise error
    if inp is None:
        return {'cat': 0.3, 'dog': 0.7}

    # preprocess
    image = preprocess_image(inp)
    image = image.to(dtype=torch.float32)

    # inference
    result = model(image)

    # postprocess
    result = torch.nn.functional.softmax(result, dim=1) # apply softmax
    result = result[0].detach().numpy().tolist() # take the first batch
    labeled_result = {name:score for name, score in zip(class_names, result)}

    return labeled_result

demo = gr.Interface(fn=image_classifier, inputs="image", outputs="label")
demo.launch(share=True)