Pre-trained GoogLeNet-128?

wangxianliang commented 8 years ago

Could you please release the pre-trained GoogLeNet-128 model? Thanks a lot!

ducha-aiki commented 8 years ago

Unfortunately I cannot find it, probably, deleted at some of the disc cleaning sessions. :(

wangxianliang commented 8 years ago

How do you prepare the lmdb data? Do you resize the image to have the same width & height?

ducha-aiki commented 8 years ago

No, I resize image to 144xN keeping the aspect ratio, where N > 144. It is done by image magick. And LMDB is used without resize flags

#!/bin/bash
#usage: ./resize_to144_trim_no_aspect.sh input_dir 
DIR=$1
FILES=`find $1/ -maxdepth 5 -name "*.*"` # Base find command
for f in $FILES
do
  if [ -f $f ]; then
   mogrify -resize "144x144^" $f
  fi
done

ducha-aiki commented 8 years ago

@wangxianliang here is fresh vgg16-128 https://github.com/ducha-aiki/caffenet-benchmark/releases/tag/vgg16_128

wangxianliang commented 8 years ago

Great! Thanks a lot.

发件人: Dmytro Mishkin [mailto:notifications@github.com] 发送时间: 2016年5月9日 18:21 收件人: ducha-aiki/caffenet-benchmark caffenet-benchmark@noreply.github.com 抄送: supercloud wangxianliang@outlook.com; Mention mention@noreply.github.com 主题: Re: [ducha-aiki/caffenet-benchmark] Pre-trained GoogLeNet-128? (#14)

@wangxianliang https://github.com/wangxianliang here is fresh vgg16-128 https://github.com/ducha-aiki/caffenet-benchmark/releases/tag/vgg16_128

— You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub https://github.com/ducha-aiki/caffenet-benchmark/issues/14#issuecomment-217828291 https://github.com/notifications/beacon/AJxnRXns7PzNsY-64-lvJYCMrLQIb5LVks5p_wqNgaJpZM4II6L9.gif

ducha-aiki / caffenet-benchmark

Pre-trained GoogLeNet-128? #14