encode_type convert_imageset.cpp

[Caleb Belth]

In the create_imagenet.sh script included with the ImageNet tutorial, the convert_imageset.cpp program in caffe/tools is not passed the optional encode_type parameter. The images for ImageNet are .JPEG, but I am now training on my own dataset. I have .png files. Do I need to pass "encode_type=png" as a parameter? I tried it both ways—first with the parameter, then without, and the generated lmdb directories are about four times larger if I leave the parameter off. What is the default encode_type if none is specified? Is there a way to look at the resulting images in the lmdb to see which looks correct?