SergioMtz commited on
Commit
1c4f5a0
1 Parent(s): 4e28f14

Primer App

Files changed (1) hide show
  1. app.py +106 -0
app.py ADDED
@@ -0,0 +1,106 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import tensorflow as tf
2
+ from tensorflow.keras.layers import Dense, Conv2D, MaxPool2D, Dropout, Add, Layer, Flatten, BatchNormalization, Activation
3
+ from tensorflow.keras.models import Model
4
+
5
+ class ResLayer(Layer):
6
+ def __init__(self, filters, name = "Res_Layer"):
7
+ super(ResLayer, self).__init__(name = name)
8
+ self.filters = filters
9
+ self.f1, self.f2, self.f3 = self.filters
10
+
11
+ # Camino normal
12
+ self.Conv_1 = Conv2D(filters = self.f1, kernel_size = (1, 1), strides = (1, 1))
13
+ self.MaxPool_1 = MaxPool2D(pool_size = (2, 2))
14
+ self.BatchNorm_1 = BatchNormalization()
15
+ self.Activation_1 = Activation("relu")
16
+
17
+ self.Conv_2 = Conv2D(filters = self.f2, kernel_size = (1, 1), strides = (1, 1))
18
+ self.BatchNorm_2 = BatchNormalization()
19
+ self.Activation_2 = Activation("relu")
20
+
21
+ self.Conv_3 = Conv2D(filters = self.f3, kernel_size = (1, 1), strides = (1, 1))
22
+ self.BatchNorm_3 = BatchNormalization()
23
+
24
+ # Camino corto
25
+ self.Conv_4 = Conv2D(filters = self.f3, kernel_size = (1, 1), strides = (1, 1))
26
+ self.MaxPool_2 = MaxPool2D(pool_size = (2, 2))
27
+
28
+ self.Add = Add()
29
+ self.Activation_3 = Activation("relu")
30
+
31
+ def call(self, inputs):
32
+ X_copy = inputs
33
+
34
+ X = self.Conv_1(inputs)
35
+ X = self.MaxPool_1(X)
36
+ X = self.BatchNorm_1(X)
37
+ X = self.Activation_1(X)
38
+
39
+ X = self.Conv_2(X)
40
+ X = self.BatchNorm_2(X)
41
+ X = self.Activation_2(X)
42
+
43
+ X = self.Conv_3(X)
44
+ X = self.BatchNorm_3(X)
45
+
46
+ X_copy = self.Conv_4(X_copy)
47
+ X_copy = self.MaxPool_2(X_copy)
48
+
49
+ outputs = self.Add([X, X_copy])
50
+ outputs = self.Activation_3(outputs)
51
+
52
+ return outputs
53
+
54
+ class ResNet(Model):
55
+ def __init__(self, filters = [[64, 128, 256]], name = "ResNet"):
56
+ super(ResNet, self).__init__(name = name)
57
+ self.filters = filters
58
+ self.nb_layers = tf.shape(self.filters)[0].numpy()
59
+ self.res_layer = [ResLayer(filters)
60
+ for i, filters in enumerate(self.filters)]
61
+ self.Flatten = Flatten()
62
+ self.Dense_1 = Dense(units = 128, activation = "relu")
63
+ self.dropout_1 = Dropout(rate = 0.2)
64
+ self.Dense_2 = Dense(units = 64, activation = "relu")
65
+ self.dropout_2 = Dropout(rate = 0.1)
66
+ self.Dense_Out = Dense(units = 10, activation = "softmax")
67
+
68
+ def call(self, inputs):
69
+ outputs = inputs
70
+ for i in range(self.nb_layers):
71
+ outputs = self.res_layer[i](outputs)
72
+
73
+ outputs = self.Flatten(outputs)
74
+ outputs = self.Dense_1(outputs)
75
+ outputs = self.dropout_1(outputs)
76
+ outputs = self.Dense_2(outputs)
77
+ outputs = self.dropout_2(outputs)
78
+ outputs = self.Dense_Out(outputs)
79
+
80
+ return outputs
81
+
82
+ model = ResNet()
83
+ model.build(input_shape = [None, 28, 28, 1])
84
+ model.load_weights("ResNet_Weights.tf")
85
+
86
+ import gradio as gr
87
+
88
+ def digit_recognition(img):
89
+ img = img / 255.
90
+ img = tf.expand_dims(img, axis = -1)
91
+ img = tf.convert_to_tensor([img], dtype = tf.float32)
92
+ prediction = model(img)
93
+ prediction = tf.squeeze(prediction)
94
+ return {"Cero": float(prediction[0]),
95
+ "Uno": float(prediction[1]),
96
+ "Dos": float(prediction[2]),
97
+ "Tres": float(prediction[3]),
98
+ "Cuatro": float(prediction[4]),
99
+ "Cinco": float(prediction[5]),
100
+ "Seis": float(prediction[6]),
101
+ "Siete": float(prediction[7]),
102
+ "Ocho": float(prediction[8]),
103
+ "Nueve": float(prediction[9])}
104
+
105
+ app = gr.Interface(fn = digit_recognition, inputs = "sketchpad", outputs = "label", description = "Dibuja un número", title = "MNIST Digit Recognition")
106
+ app.launch(share = True)