sample_script_learn.m

Standard model object recognition matlab code

% sample_script_learn.m
%
% It is an example of how to run the model on multiple images (of
% the same size).

% Define file names to intermediate, learned features.
global learn_file;
learn_file{3} = './Data/f_S2b_my_learn';
learn_file{5} = './Data/f_S2_my_learn';
learn_file{7} = './Data/f_S3_my_learn';
learn_file{9} = './Data/f_S4_my_learn';

stim_dir = './Images/'; 
save_dir = './Model_Resp/';

level = [1 1]; % Upto C2b, bypass route only, and save C2b
s_lvl = [0 1];
r_out = main_model_load_save(stim_dir,level,s_lvl,save_dir);

stim_dir = './Model_Resp/Layer2/';

% Case 1: Learn S2b
f = learn_filters_load_save(stim_dir,learn_file{3},3); 
% Case 2: Learn S2
f = learn_filters_load_save(stim_dir,learn_file{5},5);
% Case 3: Learn S3 (based on learned S2 features)
stim_dir = './Model_Resp/Layer2/';
save_dir = './Model_Resp/';
level = [0 0 0 0 1 1];
s_lvl = [0 0 0 0 1 1];
r_out = main_model_load_save(stim_dir,level,s_lvl,save_dir);
save_dir = './Model_Resp/Layer6/';
f = learn_filters_load_save(save_dir,learn_file{7},7);

% Note that the learning from a small database of images will
% create learn_file's with a small number of features, and there
% will be some conflicts with init_filter_def.m.  Therefore,
% consider augmenting the learned features with random values or
% copies of itself.  For example,

if 1
  which_learn_file = 3;
  load(learn_file{which_learn_file});
  s_tag = '_s3';

  new_num_feat = 2000; 
  old_num_feat = f.size_s3(4);

  % Take the first 400 learned features and repeat them.  Note that
  % this number may be smaller/larger depending on the size of
  % learned images.
  f1 = f.f1_s3(:,1:400);
  f2 = f.f2_s3(:,1:400);
  i1 = f.i1_s3(:,1:400);
  i2 = f.i2_s3(:,1:400);
  i3 = f.i3_s3(:,1:400);
  siz = [f.size_s3(1:3) new_num_feat];
  shift = f.shift_s3;

  f1_new = repmat(f1, [1 5]);
  f2_new = repmat(f2, [1 5]);
  i1_new = repmat(i1, [1 5]); 
  i2_new = repmat(i2, [1 5]); 
  i3_new = repmat(i3, [1 5]); 

  f_old = f;
  f_new = struct('f1_s3', f1_new, 'f2_s3', f2_new, ...
      'i1_s3', i1_new, 'i2_s3', i2_new, 'i3_s3', i3_new, ...
      'size_s3', siz, 'shift_s3', shift);
  f = f_new;
  save(learn_file{which_learn_file}, 'f');
end