ZF/OpenFace
1
0
Fork 0
mirror of https://gitcode.com/gh_mirrors/ope/OpenFace.git synced 2026-05-14 11:17:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

More work on AU - WIP

Browse Source
This commit is contained in:
Tadas Baltrusaitis
2016-06-14 17:55:16 -04:00
parent 65e91da3a6
commit 146dcd5e5b
676 changed files with 2971 additions and 297 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
matlab_version/AU_training/experiments/training_code/svr_test_linear.m
View File

@@ -8,10 +8,24 @@ function [result, prediction] = svr_test_linear(test_labels, test_samples, model
prediction(prediction>5)=5;
% using CCC as the evaluation metric
result = corr(test_labels, prediction);
[ ~, ~, ~, ccc, ~, ~ ] = evaluate_regression_results( prediction, test_labels );
result = ccc;
% using the average of CCC errors if different datasets are used
if(~isfield(model, 'eval_ids'))
result = corr(test_labels, prediction);
[ ~, ~, ~, ccc, ~, ~ ] = evaluate_regression_results( prediction, test_labels );
result = ccc;
else
eval_ids = unique(model.eval_ids)';
ccc = 0;
fprintf('CCC: ');
for i=eval_ids
[ ~, ~, ~, ccc_curr, ~, ~ ] = evaluate_regression_results( prediction(model.eval_ids == i), test_labels(model.eval_ids == i) );
ccc = ccc + ccc_curr;
fprintf('%.3f ', ccc_curr);
end
ccc = ccc / numel(eval_ids);
fprintf('mean : %.3f\n', ccc);
result = ccc;
end
if(isnan(result))
result = 0;

21
matlab_version/AU_training/experiments/training_code/svr_test_linear_shift.m
View File

@@ -29,10 +29,23 @@ function [result, prediction] = svr_test_linear_shift(test_labels, test_samples,
% using the average of RMS errors
% result = mean(sqrt(mean((prediction - test_labels).^2)));
result = corr(test_labels, prediction);
[ ~, ~, ~, ccc, ~, ~ ] = evaluate_regression_results( prediction, test_labels );
result = ccc;
if(~isfield(model, 'eval_ids'))
result = corr(test_labels, prediction);
[ ~, ~, ~, ccc, ~, ~ ] = evaluate_regression_results( prediction, test_labels );
result = ccc;
else
eval_ids = unique(model.eval_ids)';
ccc = 0;
fprintf('CCC: ');
for i=eval_ids
[ ~, ~, ~, ccc_curr, ~, ~ ] = evaluate_regression_results( prediction(model.eval_ids == i), test_labels(model.eval_ids == i) );
ccc = ccc + ccc_curr;
fprintf('%.3f ', ccc_curr);
end
ccc = ccc / numel(eval_ids);
fprintf('mean : %.3f\n', ccc);
result = ccc;
end
if(isnan(result))
result = 0;

41
matlab_version/AU_training/experiments/training_code/svr_test_linear_shift_fancy.m
View File

@@ -1,41 +0,0 @@
function [result, prediction] = svr_test_linear_shift_fancy(test_labels, test_samples, model)
prediction = test_samples * model.w(1:end-1)' + model.w(end);
% prediction = predict(test_labels, test_samples, model);
prediction(~model.success) = 0;
if(model.cutoff >= 0)
% perform shifting here per person
users = unique(model.vid_ids);
for i=1:numel(users)
preds_user = prediction(strcmp(model.vid_ids, users(i)));
sorted = sort(preds_user);
% alternative, move to histograms and pick the highest one
shift = sorted(round(end*model.cutoff)+1);
prediction(strcmp(model.vid_ids, users(i))) = preds_user - shift;
end
end
% Cap the prediction as well
prediction(prediction<0)=0;
prediction(prediction>5)=5;
% using the average of RMS errors
% result = mean(sqrt(mean((prediction - test_labels).^2)));
result = corr(test_labels, prediction);
[ ~, ~, ~, ccc, ~, ~ ] = evaluate_regression_results( prediction, test_labels );
result = ccc;
if(isnan(result))
result = 0;
end
end

5
matlab_version/AU_training/experiments/training_code/svr_train_linear.m
View File

@@ -5,4 +5,9 @@ function [model] = svr_train_linear(train_labels, train_samples, hyper)
comm = sprintf('-s 11 -B 1 -p %.10f -c %.10f -q', hyper.p, hyper.c);
model = train(train_labels, train_samples, comm);
model.success = hyper.success;
if(isfield(hyper, 'eval_ids'))
model.eval_ids = hyper.eval_ids;
end
end

7
matlab_version/AU_training/experiments/training_code/svr_train_linear_shift_fancy.m → matlab_version/AU_training/experiments/training_code/svr_train_linear_shift.m
View File

@@ -1,4 +1,4 @@
function [model] = svr_train_linear_shift_fancy(train_labels, train_samples, hyper)
function [model] = svr_train_linear_shift(train_labels, train_samples, hyper)
% Change to your downloaded location
addpath('C:\liblinear\matlab')
@@ -54,4 +54,9 @@ function [model] = svr_train_linear_shift_fancy(train_labels, train_samples, hyp
model.cutoff = cutoffs(best_id);
model.vid_ids = hyper.vid_ids;
model.success = hyper.success;
if(isfield(hyper, 'eval_ids'))
model.eval_ids = hyper.eval_ids;
end
end