Keras implementation of DeXpression architecture mentioned in this paper: https://arxiv.org/abs/1509.05371
model = DeXpression(include_top=True,
weights=None,
input_tensor=None,
input_shape=None,
padding='same',
classes=7)
model.summary()
____________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
====================================================================================================
input_1 (InputLayer) (None, 224, 224, 3) 0
____________________________________________________________________________________________________
conv_1 (Conv2D) (None, 112, 112, 64) 9472 input_1[0][0]
____________________________________________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None, 55, 55, 64) 0 conv_1[0][0]
____________________________________________________________________________________________________
batch_normalization_1 (BatchNorm (None, 55, 55, 64) 256 max_pooling2d_1[0][0]
____________________________________________________________________________________________________
conv_2a (Conv2D) (None, 55, 55, 96) 6240 batch_normalization_1[0][0]
____________________________________________________________________________________________________
maxpool_2a (MaxPooling2D) (None, 55, 55, 64) 0 batch_normalization_1[0][0]
____________________________________________________________________________________________________
conv_2b (Conv2D) (None, 55, 55, 208) 179920 conv_2a[0][0]
____________________________________________________________________________________________________
conv_2c (Conv2D) (None, 55, 55, 64) 4160 maxpool_2a[0][0]
____________________________________________________________________________________________________
concat_2 (Merge) (None, 55, 55, 272) 0 conv_2b[0][0]
conv_2c[0][0]
____________________________________________________________________________________________________
maxpool_2b (MaxPooling2D) (None, 55, 55, 272) 0 concat_2[0][0]
____________________________________________________________________________________________________
conv_3a (Conv2D) (None, 55, 55, 96) 26208 maxpool_2b[0][0]
____________________________________________________________________________________________________
conv_3b (Conv2D) (None, 55, 55, 208) 179920 conv_3a[0][0]
____________________________________________________________________________________________________
conv_3c (Conv2D) (None, 55, 55, 64) 4160 maxpool_2a[0][0]
____________________________________________________________________________________________________
concat_3 (Merge) (None, 55, 55, 272) 0 conv_3b[0][0]
conv_3c[0][0]
____________________________________________________________________________________________________
maxpool_3b (MaxPooling2D) (None, 55, 55, 272) 0 concat_3[0][0]
____________________________________________________________________________________________________
flatten_1 (Flatten) (None, 822800) 0 maxpool_3b[0][0]
____________________________________________________________________________________________________
predictions (Dense) (None, 7) 5759607 flatten_1[0][0]
====================================================================================================
Total params: 6,169,943
Trainable params: 6,169,815
Non-trainable params: 128
____________________________________________________________________________________________________