Imagenet Tutorial. create_imagenet.sh / convert_imageset.cpp

[Junho Cho]

 Hi, I am new to caffe and following the Imagenet Tutorial (http://caffe.berkeleyvision.org/gathered/examples/imagenet.html)

 And having problem on executing "create_imagenet.sh".

 This command which change the ImageNet2012 data into lmdb format takes more than 8h time and requires more than 730GB, comparing that original Imagenet data is 135GB.

 Is this normal? 

 FYI, I only resized pictures of "imagenet/val" into 256x256. Am I also supposed to resize all pictures of 135GB ImageNet data? Resizing images also take much time.

The tutorial says to use mincepie but have no idea how to use. I want to train model by myself and having troubles on consuming much time and resources.

Do you have any idea on my problems or better option to learn the process of training my own model?

Thx a lot!

[Junho Cho]

Attach some more information

2015년 6월 8일 월요일 오후 8시 16분 25초 UTC+9, Junho Cho 님의 말: