projects / platform / upstream / caffeonacl.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 824c344) | patch
replace old cifar full with within channel LRN (per cuda-convnet
authorJeff Donahue <jeff.donahue@gmail.com>
Sat, 29 Mar 2014 11:26:35 +0000 (04:26 -0700)
committerJeff Donahue <jeff.donahue@gmail.com>
Tue, 8 Apr 2014 18:36:18 +0000 (11:36 -0700)
commit24f7318ef2a50e59d2fc8862f4c7426392b9c8a7
treee62dd8d5e32a8129350ec2b352c4c08050934a86tree | snapshot
parent824c34483ec4360cd660ec9ca60561cd52da8b86commit | diff
replace old cifar full with within channel LRN (per cuda-convnet
layers-18pct) -- slightly slower (5000 iters now takes 6:57; took 6:43
previously), but slightly more accurate (exactly 82% test accuracy; got
81.65% before)
examples/cifar10/cifar10_full_lrn_in_channel_solver.prototxt [deleted file] blob | history
examples/cifar10/cifar10_full_lrn_in_channel_solver_lr1.prototxt [deleted file] blob | history
examples/cifar10/cifar10_full_lrn_in_channel_solver_lr2.prototxt [deleted file] blob | history
examples/cifar10/cifar10_full_lrn_in_channel_test.prototxt [deleted file] blob | history
examples/cifar10/cifar10_full_lrn_in_channel_train.prototxt [deleted file] blob | history
examples/cifar10/cifar10_full_test.prototxt diff | blob | history
examples/cifar10/cifar10_full_train.prototxt diff | blob | history
examples/cifar10/train_full.sh diff | blob | history
examples/cifar10/train_full_lrn_in_channel.sh [deleted file] blob | history
Domain: Machine Learning / ML Framework;