Skip to first unread message
Gary Dong
Jan 29, 2016, 9:30:54 PM1/29/16
to Caffe Users
Hi, I'm new in Caffe. I'm trying to use Caffe to train my signal data. The datasets are one dimensional time series saved in .txt files. So do I have to turn them to .jpg and use the convert_image to turn them into lmdb? Or is there any way to get them directly into the Caffe?
Jan C Peters
Jan 30, 2016, 6:04:02 AM1/30/16
to Caffe Users
Turning them into JPG is a horrible idea. Directly convert them to LMDB yourself, look at the existing scripts to see how to do that. A few days ago I posted a python script that shows how to do something similar: https://groups.google.com/forum/#!msg/caffe-users/2xpmLJYmt5k/ApiOQ7NnAwAJ
Or directly put it into HDF5 files. That would be much easier.
On a different matter: I don't think caffe is very well suited for sequence learning, don't people usually use recurrent models for that? Vanilla caffe does not really support recurrent connections (apart from loops) afaik, but there are forks for RNNs.
Jan
Or directly put it into HDF5 files. That would be much easier.
On a different matter: I don't think caffe is very well suited for sequence learning, don't people usually use recurrent models for that? Vanilla caffe does not really support recurrent connections (apart from loops) afaik, but there are forks for RNNs.
Jan
Gary Dong
Jan 30, 2016, 9:56:43 AM1/30/16
to Caffe Users
Thank you very much! It helps me a lot.
Talking about the sequence learning, I'm also considering using TensorFlow which seems to work well with RNNs.
在 2016年1月30日星期六 UTC+8下午7:04:02，Jan C Peters写道：
Talking about the sequence learning, I'm also considering using TensorFlow which seems to work well with RNNs.
在 2016年1月30日星期六 UTC+8下午7:04:02，Jan C Peters写道：
Reply all
Reply to author
Forward
0 new messages