Model-Tuner error on lenet-mnist model
tzitziris opened this issue · comments
Hello I have been trying to use the Model Tuner tool on the Lenet Mnist dataset with a image dataset of numbers(0-9) and while the previous image-classifier and model-compliler run fine, I get a Segmentation. I upload the files that I use. Any ideas where is the problem?
ampampos@sheldon:~/tries/buildwithbundles/bin$ ./model-tuner -backend=CPU -model=/home/ampampos/tries/lenettest -image-mode=0to1 -quantization-schema=asymmetric -quantization-precision=Int8 -quantization-precision-bias=Int8 -load-profile=/home/ampampos/tries/buildwithbundles/bin/profiletest5622a3.yml -dataset-path=/home/ampampos/tries/atest/Mnist_png -dataset-file=/home/ampampos/tries/atest/metadata.txt -model-input=data,float,[1,1,28,28] -dump-tuned-profile=tunerbuild5622a3.yml
Computing initial accuracy ...
WARNING: Logging before InitGoogleLogging() is written to STDERR
I0606 08:46:28.568809 628899 Partitioner.cpp:493] The model is too small for applying partition.
Model size : 1727456
Backend Name : CPU
Device memory: 2000000000
I0606 08:46:28.573915 628899 Partitioner.cpp:90] The number of partitions is : 1
I0606 08:46:28.573932 628899 PartitionerUtils.cpp:551] Partition 0:
Name : /home/ampampos/tries/lenettest
BackendKind : CPU
context count : 1
total Memory : 1727496
input size: 3136
input count : 1
input only from peers count : 0
output size: 40
constant size: 1724320
non-deferred constant size: 1724320
deferred constant size: 0
I0606 08:46:28.573966 628899 PartitionerUtils.cpp:578] LogicalDeviceIDs : 0
./model-tuner[0xc72e5e]
./model-tuner[0xc710e1]
./model-tuner[0xc73264]
/lib/x86_64-linux-gnu/libpthread.so.0(+0x1441f)[0x7f7cd45e941f]
./model-tuner[0x500686]
./model-tuner[0x501715]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf2)[0x7f7cd40b9082]
./model-tuner[0x4bab6d]
Segmentation fault (core dumped)
Finally, we found the solution, we must add the command "-image-channel-order=RGB"