-######################################################################
-## INTRODUCTION
+I. INTRODUCTION
- This is the C++ implementation of the folded hierarchy of
- classifiers for cat detection described in
+ This is the open-source C++ implementation of the folded hierarchy
+ of classifiers for cat detection described in
F. Fleuret and D. Geman, "Stationary Features and Cat Detection",
Journal of Machine Learning Research (JMLR), 2008, to appear.
- Please cite this paper when referring to this software.
+ Please use that citation when referring to this software.
-######################################################################
-## INSTALLATION
+ Contact Francois Fleuret at fleuret@idiap.ch for comments and bug
+ reports.
- This program was developed on Debian GNU/Linux computers with the
- following main tool versions
-
- * GNU bash, version 3.2.39
- * g++ 4.3.2
- * gnuplot 4.2 patchlevel 4
+II. INSTALLATION
If you have installed the RateMyKitten images provided on
- http://www.idiap.ch/folded-ctf
+ http://www.idiap.ch/folded-ctf
in the source directory, everything should work seamlessly by
- invoking the ./run.sh script. It will
+ invoking the ./run.sh script.
+
+ It will
* Compile the source code entirely
* Run 20 rounds of training / test (ten rounds for each of HB and
H+B detectors with different random seeds)
- You can also run the full thing with the following commands if you
- have wget installed
+ You can run the full thing with the following commands if you have
+ wget installed
- > wget http://www.idiap.ch/folded-ctf/not-public-yet/data/folding-gpl.tgz
+ > wget http://www.idiap.ch/folded-ctf/data/folding-gpl.tgz
> tar zxvf folding-gpl.tgz
> cd folding
- > wget http://www.idiap.ch/folded-ctf/not-public-yet/data/rmk.tgz
+ > wget http://www.idiap.ch/folded-ctf/data/rmk.tgz
> tar zxvf rmk.tgz
> ./run.sh
Note that every one of the twenty rounds of training/testing takes
more than three days on a powerful PC. However, the script detects
already running computations by looking at the presence of the
- corresponding result directory. Hence, it can be run in parallel on
- several machines as long as they see the same result directory.
+ corresponding result directories. Hence, it can be run in parallel
+ on several machines as long as they see the same result directory.
When all or some of the experimental rounds are over, you can
- generate the ROC curves by invoking the ./graph.sh script.
+ generate the ROC curves by invoking the ./graph.sh script. You need
+ a fairly recent version of Gnuplot.
+
+ This program was developed on Debian GNU/Linux computers with the
+ following main tool versions
+
+ * GNU bash, version 3.2.39
+ * g++ 4.3.2
+ * gnuplot 4.2 patchlevel 4
- You are welcome to send bug reports and comments to fleuret@idiap.ch
+ Due to approximations in the optimized arithmetic operations with
+ g++, results may vary with different versions of the compiler
+ and/or different levels of optimization.
-######################################################################
-## PARAMETERS
+III. PARAMETERS
- To set the value of a parameter during an experiment, just add an
- argument of the form --parameter-name=value before the commands that
- should take into account that value.
+ To set the value of a parameter, just add an argument of the form
+ --parameter-name=value before the commands that should take it into
+ account.
For every parameter below, the default value is given between
parenthesis.
* pool-name (no default)
- Where are the data to use
+ The scene pool file name.
* test-pool-name (no default)
- Should we use a separate pool file, and ignore proportion-for-test
- then.
+ Should we use a separate test pool file. If none is given, then
+ the test scenes are taken at random from the main pool file
+ according to proportion-for-test.
* detector-name ("default.det")
* tree-depth-max (1)
Maximum depth of the decision trees used as weak learners in the
- classifier. The default value corresponds to stumps.
+ classifier. The default value of 1 corresponds to stumps.
* proportion-negative-cells-for-training (0.025)
Overall proportion of negative cells to use during learning (we
- sample among them)
+ sample among for boosting).
* nb-negative-samples-per-positive (10)
* nb-features-for-boosting-optimization (10000)
- How many pose-indexed features to use at every step of boosting.
+ How many pose-indexed features to look at for optimization at
+ every step of boosting.
* force-head-belly-independence ("no")
Should we force the independence between the two levels of the
detector (i.e. make an H+B detector)
- * nb-weak-learners-per-classifier (10)
+ * nb-weak-learners-per-classifier (100)
- This parameter corresponds to the value U in the JMLR paper, and
- should be set to 100.
+ This parameter corresponds to the value U in the article.
* nb-classifiers-per-level (25)
- This parameter corresponds to the value B in the JMLR paper.
+ This parameter corresponds to the value B in the article.
- * nb-levels (1)
+ * nb-levels (2)
- How many levels in the hierarchy. This should be 2 for the JMLR
- paper experiments.
+ How many levels in the hierarchy.
- * proportion-for-train (0.5)
+ * proportion-for-train (0.75)
The proportion of scenes from the pool to use for training.
* write-parse-images ("no")
Should we save one image for every test scene with the resulting
- alarms.
+ alarms. This option generates a lot of images for every round and
+ is switched off by default. Switch it on to produce images such as
+ the full page of results in the paper.
* write-tag-images ("no")
Should we save the (very large) tag images when saving the
materials.
- * wanted-true-positive-rate (0.5)
+ * wanted-true-positive-rate (0.75)
What is the target true positive rate. Note that this is the rate
without post-processing and without pose tolerance in the
* progress-bar ("yes")
- Should we display a progress bar.
+ Should we display a progress bar during long computations.
-######################################################################
-## COMMANDS
+IV. COMMANDS
* open-pool
* compute-thresholds
- Compute the thresholds of the detector classifiers to obtain the
- required wanted-true-positive-rate
+ Compute the thresholds of the detector classifiers from the
+ validation set to obtain the required wanted-true-positive-rate.
* test-detector
Visit nb-wanted-true-positive-rates rates between 0 and
wanted-true-positive-rate, for each compute the detector
- thresholds on the validation set, estimate the error rate on the
- test set.
+ thresholds on the validation set and estimate the error rate on
+ the test set.
* write-detector
* write-pool-images
- Write PNG images of the scenes in the pool.
+ For every of the first nb-images of the pool, save one PNG image
+ with the ground truth, one with the corresponding referential at
+ the reference scale, and one with the feature material-feature-nb
+ from the detector. This last image is not saved if either no
+ detector has been read/trained or if no feature number has been
+ specified.
--
Francois Fleuret
void Global::init_parser(ParamParser *parser) {
// The nice level of the process
- parser->add_association("niceness", "5", false);
+ parser->add_association("niceness", "15", false);
// Seed to initialize the random generator
parser->add_association("random-seed", "0", false);
// How many images to produce/process
parser->add_association("nb-images", "-1", false);
+ // What is the number of the feature to show in the images
+ parser->add_association("material-feature-nb", "-1", false);
// What is the maximum tree depth
parser->add_association("tree-depth-max", "1", false);
// Do we allow head-belly registration
parser->add_association("force-head-belly-independence", "no", false);
// How many weak-learners in every classifier
- parser->add_association("nb-weak-learners-per-classifier", "10", false);
+ parser->add_association("nb-weak-learners-per-classifier", "100", false);
// How many classifiers per level
parser->add_association("nb-classifiers-per-level", "25", false);
// How many levels
- parser->add_association("nb-levels", "1", false);
+ parser->add_association("nb-levels", "2", false);
// Proportion of images from the pool to use for training
parser->add_association("proportion-for-train", "0.5", false);
parser->add_association("write-tag-images", "no", false);
// What is the wanted true overall positive rate
- parser->add_association("wanted-true-positive-rate", "0.5", false);
+ parser->add_association("wanted-true-positive-rate", "0.75", false);
// How many rates to try for the sequence of tests
parser->add_association("nb-wanted-true-positive-rates", "10", false);
}
nb_images = parser->get_association_int("nb-images");
+ material_feature_nb = parser->get_association_int("material-feature-nb");
tree_depth_max = parser->get_association_int("tree-depth-max");
nb_weak_learners_per_classifier = parser->get_association_int("nb-weak-learners-per-classifier");
nb_classifiers_per_level = parser->get_association_int("nb-classifiers-per-level");
char detector_name[buffer_size];
char result_path[buffer_size];
- char materials_image_numbers[buffer_size];
- char materials_pf_numbers[buffer_size];
-
int loss_type;
int nb_images;
+ int material_feature_nb;
int tree_depth_max;
referential->draw(&result_sp, level);
}
- (*global.log_stream) << "Writing " << filename << endl;
- result_sp.write_png(filename);
-}
-
-void write_one_pi_feature_png(char *filename,
- LabelledImage *image,
- PoseCellHierarchy *hierarchy,
- int nb_target,
- int level,
- PiFeature *pf) {
-
- PoseCell target_cell;
- hierarchy->get_containing_cell(image, level,
- image->get_target_pose(nb_target), &target_cell);
- PiReferential referential(&target_cell);
- RGBImage result(image->width(), image->height());
- image->to_rgb(&result);
- RGBImageSubpixel result_sp(&result);
- referential.draw(&result_sp, level);
- // pf->draw(&result_sp, 255, 255, 0, &referential);
+ cout << "Writing " << filename << endl;
result_sp.write_png(filename);
}
PoseCellHierarchy *hierarchy = new PoseCellHierarchy(pool);
- for(int i = 0; i < min(global.nb_images, pool->nb_images()); i++) {
- image = pool->grab_image(i);
- RGBImage result(image->width(), image->height());
- image->to_rgb(&result);
- RGBImageSubpixel result_sp(&result);
-
- if(global.pictures_for_article) {
- for(int t = 0; t < image->nb_targets(); t++) {
- image->get_target_pose(t)->draw(8, 255, 255, 255,
- hierarchy->nb_levels() - 1, &result_sp);
-
+ if(global.material_feature_nb < 0) {
+ for(int i = 0; i < min(global.nb_images, pool->nb_images()); i++) {
+ image = pool->grab_image(i);
+ RGBImage result(image->width(), image->height());
+ image->to_rgb(&result);
+ RGBImageSubpixel result_sp(&result);
+
+ if(global.pictures_for_article) {
+ for(int t = 0; t < image->nb_targets(); t++) {
+ image->get_target_pose(t)->draw(8, 255, 255, 255,
+ hierarchy->nb_levels() - 1, &result_sp);
+
+ }
+ for(int t = 0; t < image->nb_targets(); t++) {
+ image->get_target_pose(t)->draw(4, 0, 0, 0,
+ hierarchy->nb_levels() - 1, &result_sp);
+ }
+ } else {
+ for(int t = 0; t < image->nb_targets(); t++) {
+ image->get_target_pose(t)->draw(4, 255, 128, 0,
+ hierarchy->nb_levels() - 1, &result_sp);
+ }
}
- for(int t = 0; t < image->nb_targets(); t++) {
- image->get_target_pose(t)->draw(4, 0, 0, 0,
- hierarchy->nb_levels() - 1, &result_sp);
- }
- } else {
- for(int t = 0; t < image->nb_targets(); t++) {
- image->get_target_pose(t)->draw(4, 255, 128, 0,
- hierarchy->nb_levels() - 1, &result_sp);
- }
- }
- sprintf(buffer, "/tmp/truth-%05d.png", i);
- cout << "Writing " << buffer << endl;
- result_sp.write_png(buffer);
- pool->release_image(i);
+ sprintf(buffer, "/tmp/truth-%05d.png", i);
+ cout << "Writing " << buffer << endl;
+ result_sp.write_png(buffer);
+ pool->release_image(i);
+ }
}
for(int i = 0; i < min(global.nb_images, pool->nb_images()); i++) {
image->to_rgb(&result);
RGBImageSubpixel result_sp(&result);
- int u = 0;
-
// image->compute_rich_structure();
for(int t = 0; t < image->nb_targets(); t++) {
PiReferential referential(&target_cell);
- sprintf(buffer, "/tmp/referential-%05d-%02d.png", i, u);
image->compute_rich_structure();
- write_referential_png(buffer, hierarchy->nb_levels() - 1, image, &referential, 0);
-
- if(detector) {
- int nb_features = 100;
- for(int f = 0; f < nb_features; f++)
- if(f == 0 || f == 50 || f == 53) {
- int n_family, n_feature;
- if(f < nb_features/2) {
- n_family = 0;
- n_feature = f;
- } else {
- n_family = detector->_nb_classifiers_per_level;
- n_feature = f - nb_features/2;
- }
- pf = detector->_pi_feature_families[n_family]->get_feature(n_feature);
- sprintf(buffer, "/tmp/pf-%05d-%02d-%03d.png", i, u, f);
- write_referential_png(buffer,
- hierarchy->nb_levels() - 1,
- image,
- &referential,
- pf);
- }
+
+ if(global.material_feature_nb < 0) {
+ sprintf(buffer, "/tmp/referential-%05d-%02d.png", i, t);
+ write_referential_png(buffer, hierarchy->nb_levels() - 1, image, &referential, 0);
+ } else if(detector) {
+ int n_family = 0;
+ int n_feature = global.material_feature_nb;
+ while(n_feature > detector->_pi_feature_families[n_family]->nb_features()) {
+ n_family++;
+ n_feature -= detector->_pi_feature_families[n_family]->nb_features();
+ }
+ pf = detector->_pi_feature_families[n_family]->get_feature(n_feature);
+ sprintf(buffer, "/tmp/pf-%05d-%02d-%05d.png", i, t, global.material_feature_nb);
+ write_referential_png(buffer,
+ hierarchy->nb_levels() - 1,
+ image,
+ &referential,
+ pf);
}
- u++;
}
pool->release_image(i);
}
}
- (*global.log_stream) << "Writing " << filename << endl;
-
+ cout << "Writing " << filename << endl;
result_sp.write_png(filename);
}
void write_pool_images_with_poses_and_referentials(LabelledImagePool *pool,
Detector *detector);
-void write_one_pi_feature_png(char *filename,
- LabelledImage *image,
- PoseCellHierarchy *hierarchy,
- int nb_target,
- int level,
- PiFeature *pf);
-
void write_image_with_detections(const char *filename,
LabelledImage *image,
PoseCellSet *detections,
scalar_t belly_ryc_min = belly_resolution * floor(belly_ryc.min / belly_resolution);
int nb_belly_ryc = int(ceil((belly_ryc.max - belly_ryc_min) / belly_resolution));
- (*global.log_stream) << "belly_rxc = " << belly_rxc << endl
- << "belly_rxc_min = " << belly_rxc_min << endl
- << "belly_rxc_min + nb_belly_rxc * belly_resolution = " << belly_rxc_min + nb_belly_rxc * belly_resolution << endl
- << endl
- << "belly_ryc = " << belly_ryc << endl
- << "belly_ryc_min = " << belly_ryc_min << endl
- << "belly_ryc_min + nb_belly_ryc * belly_resolution = " << belly_ryc_min + nb_belly_ryc * belly_resolution << endl;
-
int used[nb_belly_rxc * nb_belly_rxc];
for(int k = 0; k < nb_belly_rxc * nb_belly_ryc; k++) {
_belly_cells = new RelativeBellyPoseCell[_nb_belly_cells];
- for(int j = 0; j < nb_belly_ryc; j++) {
- for(int i = 0; i < nb_belly_rxc; i++) {
- if(used[i + nb_belly_rxc * j]) {
- if(sq(scalar_t(i) * belly_resolution + belly_resolution/2 + belly_rxc_min) +
- sq(scalar_t(j) * belly_resolution + belly_resolution/2 + belly_ryc_min) <= 1) {
- (*global.log_stream) << "*";
- } else {
- (*global.log_stream) << "X";
- }
- } else {
- (*global.log_stream) << ".";
- }
- }
- (*global.log_stream) << endl;
- }
-
int k = 0;
for(int j = 0; j < nb_belly_ryc; j++) {
for(int i = 0; i < nb_belly_rxc; i++) {
}
}
}
-
- (*global.log_stream) << _nb_belly_cells << " belly cells." << endl;
}
PoseCellHierarchy::~PoseCellHierarchy() {
MAIN_URL="http://www.idiap.ch/folded-ctf"
+######################################################################
# Compiling
make -j -k
RESULT_DIR=./results
-if [[ ! -d ${RESULT_DIR} ]]; then
- mkdir ${RESULT_DIR}
-fi
+case $1 in
-for SEED in {0..9}; do
+ pics)
- for MODE in hb h+b; do
+ SEED=0
- EXPERIMENT_RESULT_DIR="${RESULT_DIR}/${MODE}-${SEED}"
+ EXPERIMENT_RESULT_DIR="${RESULT_DIR}/hb-${SEED}"
- mkdir ${EXPERIMENT_RESULT_DIR} 2> /dev/null
+ if [[ -d "${EXPERIMENT_RESULT_DIR}" ]]; then
- if [[ $? == 0 ]]; then
+ for n in -1 0 2501 2504; do
- OPTS="--random-seed=${SEED} --wanted-true-positive-rate=0.75"
-
- if [[ $MODE == "h+b" ]]; then
- OPTS="${OPTS} --force-head-belly-independence=yes"
- fi
-
- if [[ $1 == "valgrind" ]]; then
- OPTS="${OPTS} --nb-classifiers-per-level=1 --nb-weak-learners-per-classifier=10"
- OPTS="${OPTS} --proportion-for-train=0.1 --proportion-for-validation=0.025 --proportion-for-test=0.01"
- OPTS="${OPTS} --wanted-true-positive-rate=0.1"
- DEBUGGER="valgrind --db-attach=yes --leak-check=full --show-reachable=yes"
- fi
-
- ${DEBUGGER} ./folding \
- --niceness=15 \
- --pool-name=${POOL_NAME} \
- --nb-levels=2 \
- --nb-classifiers-per-level=25 --nb-weak-learners-per-classifier=100 \
- --result-path=${EXPERIMENT_RESULT_DIR} \
- --detector-name=${EXPERIMENT_RESULT_DIR}/default.det \
- ${OPTS} \
- open-pool \
- train-detector \
- compute-thresholds \
- write-detector \
- sequence-test-detector | tee -a ${EXPERIMENT_RESULT_DIR}/stdout
+ ./folding --random-seed=${SEED} \
+ --pool-name=${POOL_NAME} \
+ --result-path=${EXPERIMENT_RESULT_DIR} \
+ --detector-name=${EXPERIMENT_RESULT_DIR}/default.det \
+ --nb-images=1 \
+ --material-feature-nb=${n} \
+ open-pool \
+ read-detector \
+ write-pool-images
+
+ done
else
+ echo "You have to run at least the first round completely to be able" >&2
+ echo "to generate the pictures." >&2
+ exit 1
+ fi
+
+ ;;
- echo "${EXPERIMENT_RESULT_DIR} exists, aborting experiment."
+ valgrind|"")
+ if [[ ! -d ${RESULT_DIR} ]]; then
+ mkdir ${RESULT_DIR}
fi
- done
+ for SEED in {0..9}; do
+
+ for MODE in hb h+b; do
+
+ EXPERIMENT_RESULT_DIR="${RESULT_DIR}/${MODE}-${SEED}"
+
+ mkdir ${EXPERIMENT_RESULT_DIR} 2> /dev/null
+
+ if [[ $? == 0 ]]; then
+
+ if [[ $MODE == "h+b" ]]; then
+ OPTS="${OPTS} --force-head-belly-independence=yes"
+ fi
+
+ if [[ $1 == "valgrind" ]]; then
+ OPTS="${OPTS} --nb-classifiers-per-level=1 --nb-weak-learners-per-classifier=10"
+ OPTS="${OPTS} --proportion-for-train=0.1 --proportion-for-validation=0.025 --proportion-for-test=0.01"
+ OPTS="${OPTS} --wanted-true-positive-rate=0.1"
+ DEBUGGER="valgrind --db-attach=yes --leak-check=full --show-reachable=yes"
+ fi
+
+ ${DEBUGGER} ./folding \
+ --random-seed=${SEED} \
+ --pool-name=${POOL_NAME} \
+ --result-path=${EXPERIMENT_RESULT_DIR} \
+ --detector-name=${EXPERIMENT_RESULT_DIR}/default.det \
+ ${OPTS} \
+ open-pool \
+ train-detector \
+ compute-thresholds \
+ write-detector \
+ sequence-test-detector | tee -a ${EXPERIMENT_RESULT_DIR}/stdout
+
+ else
+
+ echo "${EXPERIMENT_RESULT_DIR} exists, aborting experiment."
+
+ fi
+
+ done
+
+ done
-done
+esac
projects / folded-ctf.git / commitdiff