-
Notifications
You must be signed in to change notification settings - Fork 18.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
MNIST demo with lmdb (addresses issues 922 and 958) #1008
Conversation
@@ -50,7 +50,8 @@ net_param { | |||
top: "data" | |||
top: "label" | |||
data_param { | |||
source: "examples/mnist/mnist_train_leveldb" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Update examples/mnist/lenet_train_test.prototxt for the lmdb source too.
Thanks for resolving the consolidated LeNet issue, fixing the MNIST db script bug, and the error message polish. p.s. Good to see your first PR -- hope to see you at ECCV! |
…nist.sh), switched around includes
…ed in convert_mnist_data.cpp to lmdb
ee9b2cf
to
5615bf6
Compare
Rebased, squashed and added the requested change. (It seems Travis CI had problems retrieving some packages. That particular commit only changes a bash script and a network definition.) P.S. Yup, I'll be there, looking forward to seeing you! |
For the record, this issue with leveldb is not exclusive to OSX (#522 (comment)), I had the same problem on Debian (wheezy). |
convert MNIST demo to lmdb, fixes
convert MNIST demo to lmdb, fixes
convert MNIST demo to lmdb, fixes
The MNIST demo now supports both lmdb (default) and leveldb. The choice of backend can be set in create_mnist.sh as a command line parameter for convert_mnist_data.bin.
Writing examples individually to the lmdb backend is literally 100x slower than using batches, so now convert_mnist_data.cpp resembles convert_imageset.cpp a lot more closely in that batch writes are used for leveldb and lmdb.
./examples/mnist/train_lenet_consolidated.sh runs to completion. The final accuracy is 99.79% and 99.2% on the training and test sets respectively.
Commit 750f92e among other things removes a bug that is currently in the tip of the dev branch.