From 9be8896b22cbbaeaea265a42275c23c05d0c839c Mon Sep 17 00:00:00 2001
From: Francois Vieille <vieille.francois@gmail.com>
Date: Mon, 11 Mar 2019 22:36:58 +0100
Subject: [PATCH] ajout d'un modele a convolution

---
 .gitignore                          |  3 +-
 iss/models/SimpleConvAutoEncoder.py | 48 +++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 1 deletion(-)
 create mode 100644 iss/models/SimpleConvAutoEncoder.py

diff --git a/.gitignore b/.gitignore
index 3f853b5..da6bdc9 100644
--- a/.gitignore
+++ b/.gitignore
@@ -92,4 +92,5 @@ config/*
 !config/config.template.yaml
 
 # models dir
-models/
\ No newline at end of file
+models/
+!iss/models
\ No newline at end of file
diff --git a/iss/models/SimpleConvAutoEncoder.py b/iss/models/SimpleConvAutoEncoder.py
new file mode 100644
index 0000000..a40d5ba
--- /dev/null
+++ b/iss/models/SimpleConvAutoEncoder.py
@@ -0,0 +1,48 @@
+# -*- coding: utf-8 -*-
+
+from iss.models.AbstractModel import AbstractModel
+from keras.layers import Input, Dense, Conv2D, MaxPooling2D, UpSampling2D, Reshape, Flatten
+from keras.optimizers import Adadelta, Adam
+from keras.models import Model
+import numpy as np
+
+class SimpleConvAutoEncoder(AbstractModel):
+
+		def __init__(self, config):
+
+			save_directory = config['save_directory']
+			model_name = config['model_name']
+
+			super().__init__(save_directory, model_name)
+
+			self.activation = config['activation']
+			self.input_shape = (config['input_height'], config['input_width'], config['input_channel'])
+			self.lr = config['learning_rate']
+			self.build_model()
+
+		def build_model(self):
+			input_shape = self.input_shape
+
+			picture = Input(shape = input_shape)
+
+			# encoded network
+			x = Conv2D(4, (3, 3), activation = 'relu', padding = 'same', name = 'enc_conv_1')(picture)
+			x = MaxPooling2D((2, 2))(x)
+			x = Conv2D(8, (3, 3), activation = 'relu', padding = 'same', name = 'enc_conv_2')(x)
+			encoded = MaxPooling2D((2, 2))(x)
+
+			# decoded network
+			x = Conv2D(8, (3, 3), activation = 'relu', padding = 'same', name = 'dec_conv_1')(encoded)
+			x = UpSampling2D((2, 2))(x)
+			x = Conv2D(4, (3, 3), activation = 'relu', padding = 'same', name = 'dec_conv_2')(x)
+			x = UpSampling2D((2, 2))(x)
+			x = Flatten()(x)
+			x = Dense(np.prod(input_shape), activation = self.activation)(x)
+			decoded = Reshape((input_shape))(x)
+
+			self.model = Model(picture, decoded)
+			
+			# optimizer = Adadelta(lr = self.lr, rho = 0.95, epsilon = None, decay = 0.0)
+			optimizer = Adam(lr = 0.001, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)
+			
+			self.model.compile(optimizer = optimizer, loss = 'binary_crossentropy')