0

I am trying to train a model with my own data though it starts training after a few iterations it throws an error.This is what it looks like 

0803 13:41:12.052456  3772 solver.cpp:237]     Train net output #2: 
loss = 1.82591 (* 1 = 1.82591 loss)
I0803 13:41:12.052477  3772 sgd_solver.cpp:105] Iteration 120, lr = 
9.92e -05
F0803 13:41:18.020836  3779 data_transformer.cpp:168] Check failed: 
height <= datum_height (227 vs. 224) 
*** Check failure stack trace: ***
    @     0x7f8f7b14adaa  (unknown)
    @     0x7f8f7b14ace4  (unknown)
    @     0x7f8f7b14a6e6  (unknown)
    @     0x7f8f7b14d687  (unknown)
    @     0x7f8f7b7ba9e0  caffe::DataTransformer<>::Transform()
    @     0x7f8f7b835a2f  caffe::DataLayer<>::load_batch()
    @     0x7f8f7b8c6a5c  
caffe::BasePrefetchingDataLayer<>::InternalThreadEntry()
    @     0x7f8f7b79aa30  caffe::InternalThread::entry()
    @     0x7f8f7b79b376  boost::detail::thread_data<>::run()
    @     0x7f8f7172aa4a  (unknown)
    @     0x7f8f6bd73184  start_thread
    @     0x7f8f79a7dffd  (unknown)
    @              (nil)  (unknown)
Aborted (core dumped)

I do think that the problem may be with input images,any suggestions as to what i should change or edit?

Ryan
  • 8,459
  • 14
  • 40
  • 66

1 Answers1

0

You probably have multiple images in a batch but they have varying sizes? You need to standardize the image sizes. You could use OpenCV to do this before feeding your images into Caffe.

arnold
  • 106
  • 2