Segmentation fault while running my own images on caffe

[Aritra Chowdhury]

Hi all,

        I am using caffe to train my own set of images on caffe. I am using only CPU on Mac OSX Yosemite. I just want to test whether the software runs properly so I have a training set of 11 images and validation set of 9 images. I have two classes labelled 0 and 1. I created the lmdb files and mean image using the shellscripts already present in caffe. However, I am getting the following error whenever I run it:

I0820 09:35:22.946832 2033701632 solver.cpp:46] Solver scaffolding done.

I0820 09:35:22.947255 2033701632 solver.cpp:237] Solving CaffeNet

I0820 09:35:22.947334 2033701632 solver.cpp:238] Learning Rate Policy: step

I0820 09:35:23.352975 2033701632 solver.cpp:281] Iteration 0, Testing net (#0)

I0820 09:35:47.902312 2033701632 solver.cpp:330]     Test net output #0: accuracy = 0

I0820 09:35:47.902365 2033701632 solver.cpp:330]     Test net output #1: loss = 7.29776 (* 1 = 7.29776 loss)

*** Aborted at 1440077762 (unix time) try "date -d @1440077762" if you are using GNU date ***

PC: @        0x108a7bd9b .L4_49

*** SIGSEGV (@0x2019938d000) received by PID 806 (TID 0x7fff7937d300) stack trace: ***

    @     0x7fff88c9ff1a _sigtramp

    @         0x545c7b50 (unknown)

Segmentation fault: 11

I went through some posts where this problem was solved by just renaming the labels. But I can't make head or tails of this problem. Please help. 

Thanks,

Aritra

[Manuel Rodriguez]

Am Donnerstag, 20. August 2015 15:50:31 UTC+2 schrieb Aritra Chowdhury:
> I am getting the following error whenever I run it:

> *** Aborted at 1440077762 (unix time) try "date -d @1440077762"

That is the logfile of the trainingphase while running caffe. Your lmdb
file is correct, because it is nearly impossible to make an error with
that. Probably your "solver.prototxt" or your "train_test.prototxt" is
buggy. The easy way to find the error, is to use the basic example of
Lenet-5 MNIST (example folder of caffe). That mean: not to use Googlenet
nor Caffenet-deploy, only Lenet-5 works. There you can adjust the path to
your data, the pixelsize of your images, the color (change from 1 to 3)
and the number of output neurons (change from 10 down to 2, because you
have only 2 classes). Hopes that works.