Try autoencoder

[yi zhao]

I am a beginner of Caffe, and I attempted to make a network of autoencoder in order to extract features from the data used in my research.
I found an example of autoencoder at "caffe/examples/mnist/mnist_autoencoder.prototxt" and used this for my network as a test run.

I prepared a small number of training image data and created database(lmdb) with "build/tools/convert_imageset".
I ran the autoencoder, houwever, it stopped with the following error message that
Check failed: bottom[0]->count() == bottom[1]->count() (3920 vs. 15750) SIGMOID_CROSS_ENTROPY_LOSS layer inputs must have the same count.

Could you tell my what is wrong and what I should change?
I have no idea because I just ran the almost the network with my own data.



My own data are a set of 20 images of size about 42 x 25.
I transformed them into 28 x 28 while creating the database(lmdb).
I rewrite only the number of batch_size from original 100 to 5 on the "caffe/examples/mnist/mnist_autoencoder.prototxt" since I used just 20 images.