 |
ML Newsnew | past | best | rss | submit | about |
| A guide to convolution arithmetic for deep learning (arxiv.org) | |
| 3 points by j_juggernaut on May 31, 2017 | hide | past | pdf | discuss on HN | |
|
In plain words: A reference that works out how a layer's input size, filter size, padding, and step size set its output size, for regular, pooling, and upsampling layers. Instead of guessing and testing, you can plug numbers into the derived formulas to size a network correctly. AbstractWe introduce a guide to help deep learning practitioners understand and manipulate convolutional neural network architectures. The guide clarifies the relationship between various properties (input shape, kernel shape, zero padding, strides and output shape) of convolutional, pooling and transposed convolutional layers, as well as the relationship between convolutional and transposed convolutional layers. Relationships are derived for various cases, and are illustrated in order to make them intuitive. | |
| add comment on HN |