deepsemantic/image_captioning

multi_crop

Closed this issue · 3 comments

Thanks for sharing the code!
But there was a question when I was training the model using “.sh” file. I aroused an error as "caffe.TransformationParameter" has no field named "multi_crop" in parsing the "multi_Bi_LSTM.prototxt", I wonder how you added the field "multi_crop. Is it because the version of CAFFE?

Hi thanks for questions. I am not sure is it the issue of different Caffe Version. I implemented it in "data_transformer.cpp" ` const int crop_size = param_.crop_size();

const Dtype scale = param_.scale();

//const bool do_mirror = param_.mirror() && Rand(2);

const bool do_mirror_x = param_.mirror() && Rand(2); // random mirror, true or false

const bool do_mirror_y = param_.mirror_vertical()&& Rand(2); // random mirror_vertical, false

const bool multi_crop=param_.multi_crop(); // if set to true, crop image from 1 of 4 corners or center

const bool multi_scale=param_.multi_scale();// if set to true, resalce image to different scale`

does some solved this issue?

@lovejasmine i have the same problem, have you solved?