rectangle make_random_cropping_rect_resnet(
const matrix<rgb_pixel>& img,
dlib::rand& rnd
)
{
// figure out what rectangle we want to crop from the image
double mins = 0.466666666, maxs = 0.875;
auto scale = mins + rnd.get_random_double()*(maxs-mins);
auto size = scale*std::min(img.nr(), img.nc());
rectangle rect(size, size);
// randomly shift the box around
point offset(rnd.get_random_32bit_number()%(img.nc()-rect.width()),
rnd.get_random_32bit_number()%(img.nr()-rect.height()));
return move_rect(rect, offset);
}
void gigaword_doc_to_vects (
const std::map<std::string, unsigned long>& words,
const gigaword_document& doc,
sparse_vector_type& lhs,
sparse_vector_type& rhs,
dlib::rand& rnd
)
{
lhs.clear();
rhs.clear();
istringstream sin(doc.text + " " + doc.headline);
unigram_tokenizer tok(sin);
string word;
while (tok(word))
{
const unsigned long id = get_word_id(words, word);
if (rnd.get_random_float() > 0.5)
{
lhs.push_back(make_pair(id, 1));
}
else
{
rhs.push_back(make_pair(id, 1));
}
}
make_sparse_vector_inplace(lhs);
make_sparse_vector_inplace(rhs);
}
void randomly_crop_images (
const matrix<rgb_pixel>& img,
dlib::array<matrix<rgb_pixel>>& crops,
dlib::rand& rnd,
long num_crops
)
{
std::vector<chip_details> dets;
for (long i = 0; i < num_crops; ++i)
{
auto rect = make_random_cropping_rect_resnet(img, rnd);
dets.push_back(chip_details(rect, chip_dims(227,227)));
}
extract_image_chips(img, dets, crops);
for (auto&& img : crops)
{
// Also randomly flip the image
if (rnd.get_random_double() > 0.5)
img = fliplr(img);
// And then randomly adjust the colors.
apply_random_color_offset(img, rnd);
}
}
void sample_hmm (
dlib::rand& rnd,
const matrix<double>& transition_probabilities,
const matrix<double>& emission_probabilities,
unsigned long previous_label,
unsigned long& next_label,
unsigned long& next_sample
)
/*!
requires
- previous_label < transition_probabilities.nr()
- transition_probabilities.nr() == transition_probabilities.nc()
- transition_probabilities.nr() == emission_probabilities.nr()
- The rows of transition_probabilities and emission_probabilities must sum to 1.
(i.e. sum_cols(transition_probabilities) and sum_cols(emission_probabilities)
must evaluate to vectors of all 1s.)
ensures
- This function randomly samples the HMM defined by transition_probabilities
and emission_probabilities assuming that the previous hidden state
was previous_label.
- The HMM is defined by:
- P(next_label |previous_label) == transition_probabilities(previous_label, next_label)
- P(next_sample|next_label) == emission_probabilities (next_label, next_sample)
- #next_label == the sampled value of the hidden state
- #next_sample == the sampled value of the observed state
!*/
{
// sample next_label
double p = rnd.get_random_double();
for (long c = 0; p >= 0 && c < transition_probabilities.nc(); ++c)
{
next_label = c;
p -= transition_probabilities(previous_label, c);
}
// now sample next_sample
p = rnd.get_random_double();
for (long c = 0; p >= 0 && c < emission_probabilities.nc(); ++c)
{
next_sample = c;
p -= emission_probabilities(next_label, c);
}
}
void randomly_crop_image (
const matrix<rgb_pixel>& img,
matrix<rgb_pixel>& crop,
dlib::rand& rnd
)
{
auto rect = make_random_cropping_rect_resnet(img, rnd);
// now crop it out as a 227x227 image.
extract_image_chip(img, chip_details(rect, chip_dims(227,227)), crop);
// Also randomly flip the image
if (rnd.get_random_double() > 0.5)
crop = fliplr(crop);
// And then randomly adjust the colors.
apply_random_color_offset(crop, rnd);
}
