image denoising convolutional autoencoder neural nets
i had an MSE of 0.02 after 10 epochs
from kaggle.com https://www.kaggle.com/adityachandrasekhar/image-super-resolution which contains RGB images of shape 256,256,3 with noise images as features and denoised one as label
Model: "model_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_2 (InputLayer) [(None, 256, 256, 3)] 0
_________________________________________________________________
conv2d_6 (Conv2D) (None, 256, 256, 256) 7168
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 128, 128, 256) 0
_________________________________________________________________
conv2d_7 (Conv2D) (None, 128, 128, 128) 295040
_________________________________________________________________
max_pooling2d_6 (MaxPooling2 (None, 64, 64, 128) 0
_________________________________________________________________
conv2d_8 (Conv2D) (None, 64, 64, 64) 73792
_________________________________________________________________
max_pooling2d_7 (MaxPooling2 (None, 32, 32, 64) 0
_________________________________________________________________
conv2d_9 (Conv2D) (None, 32, 32, 32) 18464
_________________________________________________________________
max_pooling2d_8 (MaxPooling2 (None, 16, 16, 32) 0
_________________________________________________________________
conv2d_10 (Conv2D) (None, 16, 16, 32) 9248
_________________________________________________________________
max_pooling2d_9 (MaxPooling2 (None, 8, 8, 32) 0
_________________________________________________________________
conv2d_transpose_5 (Conv2DTr (None, 16, 16, 32) 9248
_________________________________________________________________
conv2d_transpose_6 (Conv2DTr (None, 32, 32, 32) 9248
_________________________________________________________________
conv2d_transpose_7 (Conv2DTr (None, 64, 64, 64) 18496
_________________________________________________________________
conv2d_transpose_8 (Conv2DTr (None, 128, 128, 128) 73856
_________________________________________________________________
conv2d_transpose_9 (Conv2DTr (None, 256, 256, 256) 295168
_________________________________________________________________
conv2d_11 (Conv2D) (None, 256, 256, 1) 2305
=================================================================
Total params: 812,033
Trainable params: 812,033
Non-trainable params: 0