define AlexNet architecture
authorEvan Shelhamer <shelhamer@imaginarynumber.net>
Tue, 8 Apr 2014 00:30:34 +0000 (17:30 -0700)
committerEvan Shelhamer <shelhamer@imaginarynumber.net>
Tue, 15 Apr 2014 18:17:55 +0000 (12:17 -0600)
commite1b32cac3ff3f88a550d63044fcb50765d297e20
treea47d4b959cb2777b3251706c3df535ea0c70d9f4
parentb5a86710f023f351d9c692d818a22b78cade2871
define AlexNet architecture

Swap order of lrn and max pooling layers to reflect Krizhevsky
architecture. Initialize biases to 0.1 where Krizhevsky had 1, as
training by initialization with 1 has flat loss.
examples/imagenet/alexnet_deploy.prototxt [new file with mode: 0644]
examples/imagenet/alexnet_solver.prototxt [new file with mode: 0644]
examples/imagenet/alexnet_train.prototxt [new file with mode: 0644]
examples/imagenet/alexnet_val.prototxt [new file with mode: 0644]
examples/imagenet/train_alexnet.sh [new file with mode: 0755]