Using Caffe For Serial Sensor Data

[Paul Anthony Creaser]

I've just started learning and experimenting with caffe. Everything is going quite smoothly. Caffe seems to be aimed at image processing, which is ideal for the applications I am currently interested in (Road sign recognition etc...).

However I am curious, one potential use of Deep Networks, I am considering, is using serial data from sensors, such as accelerometers, gyros, pressure sensors etc... and using a Deep Network to identify difference motion types, walking, running etc...

In the theory any data type should be acceptable, it should just be a case of providing the data in a suitable format which can be processed by the network. 

[Jan C Peters]

In theory, yes. Although for sensor data, which is by nature sequential data a sequence learning approach (such as RNN or LSTM) approach is probably better suited (if you are interested in sequence properties). ON the other hand, if your signal is a sequence of images, feeding those to a CNN for instance to recognize stuff is a very valid approach. A sequence leanring approach would then even enable you to track that stuff (in theory), but I am not sure how feasible that idea is (mainly in terms of memory consumption, training time and required amount of training data).

Jan

[Evan Shelhamer]

if your signal is a sequence of images, feeding those to a CNN for instance to recognize stuff is a very valid approach. A sequence leanring approach would then even enable you to track that stuff (in theory), but I am not sure how feasible that idea is (mainly in terms of memory consumption, training time and required amount of training data).

​For one approach to learning visual sequence tasks, like recognizing activity from video, you can take a look at Long-Term Recurrent Convolutional Networks (implemented in a Caffe branch):  http://jeffdonahue.com/lrcn/

This site also has a pointer to the recurrent branch of Caffe, that handles RNN and LSTM sequence models that could be used for text, time series, and other sequence problems.​

 

Evan Shelhamer

--
You received this message because you are subscribed to the Google Groups "Caffe Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/bc2cdbd7-cc53-4e17-b8c7-7e3c34846ea6%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.