Kaggle Denoising Dirty Documents
October 10, 2015 ยท View on GitHub
My solution in this Kaggle competition "Denoising Dirty Documents", 6th place.
Model
Deep Convolutional Neural Network
| Layer Type | #Input Channel | #Ouput Channel | FilterSize | Padding |
|---|---|---|---|---|
| Input | - | 1 | - | - |
| Convolution(1) | 1 | 96 | 3 | 1 |
| Leaky ReLU | 96 | 96 | - | - |
| Convolution(2) | 96 | 96 | 1 | 0 |
| Leaky ReLU | 96 | 96 | - | - |
| Convolution(3) | 96 | 96 | 3 | 1 |
| Leaky ReLU | 96 | 96 | - | - |
| Convolution(4) | 96 | 96 | 1 | 0 |
| Leaky ReLU | 96 | 96 | - | - |
| Convolution(5) | 96 | 96 | 3 | 1 |
| Leaky ReLU | 96 | 96 | - | - |
| Convolution(6) | 96 | 1 | 3 | 1 |
| Output | 1 | - | - | - |
Sample
Test Input
Test Output
