search

BVLC / caffe

Caffe: a fast open framework for deep learning.
http://caffe.berkeleyvision.org/
Other
34.11k stars 18.7k forks source link

mnist_autoencoder.prototxt lock bug #3373

Open techmatt opened 8 years ago

techmatt commented 8 years ago

/caffe/examples/mnist/mnist_autoencoder.prototxt has two data layers, each with source: "examples/mnist/mnist-train-leveldb". On my system, caffe will hang in a deadlock when creating the data layer. I fixed the problem by duplicating the training directory, so the data layers point to different directories: "examples/mnist/mnist-train-leveldb" and "examples/mnist/mnist-train2-leveldb"

But I feel this is not a very elegant solution.

paulfitz commented 8 years ago

I see this also (lock on second reference to examples/mnist/mnist_train_lmdb).

paulfitz commented 8 years ago

Presumably same as this issue https://github.com/BVLC/caffe/issues/3037#issuecomment-140755010 with discussion about solving at #3108

astorfi commented 8 years ago

Thank you so much @techmatt , your suggestion solved my problem .... you said "But I feel this is not a very elegant solution." ... It is an elegant way used by you if someone does not want to change the "mnist_autoencoder.prototxt" file ... I try to solve it by changing and modifying the file "mnist_autoencoder.prototxt" because the problem is that, they are using so many same "data" name for bottom and top parts and for the Flatten layer the crash occurs ... Even correcting the label for the Flatten layer won't solve the problem and again it will crash in "encode1" which of type "InnerProduct"...