def detect(ibs, gpath_list, tree_path_list, **kwargs):
"""
Args:
gpath_list (list of str): the list of image paths that need detection
tree_path_list (list of str): the list of trees to load for detection
Kwargs (optional): refer to the PyRF documentation for configuration settings
Returns:
iter
"""
# Get scales from detect config, if not specified
if 'scale_list' not in kwargs.keys():
kwargs['scale_list'] = list(
map(float, ibs.cfg.detect_cfg.scale_list.split(',')))
assert all(
[isinstance(scale, float) for scale in kwargs['scale_list']])
verbose = kwargs.get('verbose', ut.VERBOSE)
if verbose:
print(
'[randomforest.detect()] Detecting with %d trees with scale_list=%r'
% (
len(tree_path_list),
kwargs['scale_list'],
))
# Run detection
detector = pyrf.Random_Forest_Detector(verbose=verbose)
forest = detector.forest(tree_path_list)
results_iter = detector.detect(forest, gpath_list, **kwargs)
return results_iter
def train_gpath_list(ibs,
train_pos_cpath_list,
train_neg_cpath_list,
trees_path=None,
**kwargs):
"""
Args:
train_pos_cpath_list (list of str): the list of positive image paths
for training
train_neg_cpath_list (list of str): the list of negative image paths
for training
trees_path (str): the path that the trees will be saved into (along
with temporary training inventory folders that are deleted once
training is finished)
species (str, optional): the species that should be used to assign to
the newly trained trees
Kwargs (optional): refer to the PyRF documentation for configuration settings
Returns:
None
"""
if trees_path is None:
trees_path = join(ibs.get_treesdir(), 'generic')
# Train trees
detector = pyrf.Random_Forest_Detector()
detector.train(train_pos_cpath_list, train_neg_cpath_list, trees_path,
**kwargs)
def compute_forgroundness(fpath1, kpts1, species='zebra_plains'):
"""
hack in foregroundness
"""
import pyrf
import vtool as vt
from os.path import exists
# hack for getting a model (not entirely ibeis independent)
trees_path = ut.get_app_resource_dir('ibeis', 'detectmodels', 'rf',
species)
tree_fpath_list = ut.glob(trees_path, '*.txt')
detector = pyrf.Random_Forest_Detector()
# TODO; might need to downsample
forest = detector.forest(tree_fpath_list, verbose=False)
gpath_list = [fpath1]
output_gpath_list = [
gpath + '.' + species + '.probchip.png' for gpath in gpath_list
]
detectkw = {
'scale_list': [1.15, 1.0, 0.85, 0.7, 0.55, 0.4, 0.25, 0.1],
'output_gpath_list': output_gpath_list,
'mode': 1, # mode one outputs probimage
}
results_iter = detector.detect(forest, gpath_list, **detectkw)
results_list = list(results_iter) # NOQA
probchip_list = [
vt.imread(gpath, grayscale=True) if exists(gpath) else None
for gpath in output_gpath_list
]
#vtpatch.get_warped_patches()
fgweights_list = []
kpts_list = [kpts1]
for probchip, kpts in zip(probchip_list, kpts_list):
patch_list = [
vt.get_warped_patch(probchip, kp)[0].astype(np.float32) / 255.0
for kp in kpts
]
weight_list = [
vt.gaussian_average_patch(patch) for patch in patch_list
]
#weight_list = [patch.sum() / (patch.size) for patch in patch_list]
weights = np.array(weight_list, dtype=np.float32)
fgweights_list.append(weights)
fgweights = fgweights_list[0]
detector.free_forest(forest)
return fgweights
