These are chat archives for beniz/deepdetect

1st
May 2017
bflick
@bflick
May 01 2017 18:25
I didn't set up any gpu support. Running in virtual box. Maybe it is trying gpu support anyway? My machine has one, but I didn't configure virtual box to use it.
It's running Ubuntu 16.04
Emmanuel Benazera
@beniz
May 01 2017 19:00
You should post the build and test output to a gist...
cchadowitz-pf
@cchadowitz-pf
May 01 2017 20:08
Hi @beniz - I'm looking to specify some transform_params on the Memory Data layer for a Caffe model and ran across (through a variety of threads/comments) the PR you have here: BVLC/caffe#5558, and related, the mdl_reset branch on beniz/caffe here: https://github.com/beniz/caffe/tree/mdl_reset
It looks like those changes are exactly what I need in order to specify mean values per channel in the deploy.prototxt instead of creating a binaryproto file instead.
However, I also realized the DeepDetect API page describes a mean param for Caffe in the Image InputConnector. Does this param do essentially the same thing? Is it R-G-B ordered?
Emmanuel Benazera
@beniz
May 01 2017 20:11
it's channeled ordered, and thus model dependent.
cchadowitz-pf
@cchadowitz-pf
May 01 2017 20:14
I see, so basically just depends on how the model was trained/deploy.prototxt is defined. Thanks!