diff options
| -rw-r--r-- | docs/tutorial/layers.md | 11 |
1 files changed, 10 insertions, 1 deletions
diff --git a/docs/tutorial/layers.md b/docs/tutorial/layers.md index 9ac3913d..fb6f2618 100644 --- a/docs/tutorial/layers.md +++ b/docs/tutorial/layers.md @@ -96,7 +96,16 @@ The `CONVOLUTION` layer convolves the input image with a set of learnable filter #### Local Response Normalization (LRN) -`LRN` +* LayerType: `LRN` +* CPU Implementation: `./src/caffe/layers/lrn_layer.cpp` +* CUDA GPU Implementation: `./src/caffe/layers/lrn_layer.cu` +* Options (`LRNParameter lrn_param`) + - Optional (default 5): `local_size`, the number of channels to sum over (for cross channel LRN) or the side length of the square region to sum over (for within channel LRN) + - Optional (default 1): `alpha`, the scaling parameter (see below) + - Optional (default 5): `beta`, the exponent (see below) + - Optional (default `ACROSS_CHANNELS`): `norm_region`, whether to sum over adjacent channels (`ACROSS_CHANNELS`) or nearby spatial locaitons (`WITHIN_CHANNEL`) + +The local response normalization layer performs a kind of "lateral inhibition" by normalizing over local input regions. In `ACROSS_CHANNELS` mode, the local regions extend across nearby channels, but have no spatial extent (i.e., they have shape `local_size x 1 x 1`). In `WITHIN_CHANNEL` mode, the local regions extend spatially, but are in separate channels (i.e., they have shape `1 x local_size x local_size`). Each input value is divided by $$(1 + (\alpha/n) \sum_i x_i)^\beta$$, where $$n$$ is the size of each local region, and the sum is taken over the region centered at that value (zero padding is added where necessary). #### im2col |