def __init__(self):
self.d = Dataset('full_pascal_trainval')
self.d_val = Dataset('full_pascal_test')
self.cls = 'dog'
suffix = 'default'
self.csc = CSCClassifier(suffix, self.cls, self.d, self.d_val)
csc_test = np.load(config.get_ext_dets_filename(self.d, 'csc_default'))
self.dets = csc_test[()]
def __init__(self):
self.d = Dataset('full_pascal_trainval')
self.d_val = Dataset('full_pascal_test')
self.cls = 'dog'
suffix = 'default'
self.csc = CSCClassifier(suffix, self.cls, self.d, self.d_val)
csc_test = np.load(config.get_ext_dets_filename(self.d, 'csc_default'))
self.dets = csc_test[()]
def load_ext_detections(cls,dataset,suffix,force=False):
"""
Loads multi-image, multi-class array of detections for all images in the
given dataset.
Loads from canonical cache location.
"""
t = time.time()
filename = config.get_ext_dets_filename(dataset, suffix)
# check for cached full file
if os.path.exists(filename) and not force:
all_dets_table = np.load(filename)[()]
else:
# TODO also return times, or process them to add to dets?
all_dets = []
for i in range(comm_rank,len(dataset.images),comm_size):
image = dataset.images[i]
if re.search('dpm',suffix):
# NOTE: not actually using the given suffix in the call below
dets = cls.load_dpm_dets_for_image(image,dataset)
ind_vector = np.ones((np.shape(dets)[0],1)) * i
dets = np.hstack((dets,ind_vector))
cols = ['x','y','w','h','dummy','dummy','dummy','dummy','score','time','cls_ind','img_ind']
good_ind = [0,1,2,3,8,9,10,11]
dets = dets[:,good_ind]
elif re.search('csc',suffix):
# NOTE: not actually using the given suffix in the call below
dets = cls.load_csc_dpm_dets_for_image(image,dataset)
ind_vector = np.ones((np.shape(dets)[0],1)) * i
dets = np.hstack((dets,ind_vector))
elif re.search('ctf',suffix):
# Split the suffix into ctf and the main part
actual_suffix = suffix.split('_')[1]
dets = cls.load_ctf_dets_for_image(image, dataset, actual_suffix)
ind_vector = np.ones((np.shape(dets)[0],1)) * i
dets = np.hstack((dets,ind_vector))
else:
print(suffix)
raise RuntimeError('Unknown detector type in suffix')
all_dets.append(dets)
final_dets = None
if comm_rank==0:
final_dets = []
safebarrier(comm)
final_dets = comm.reduce(all_dets,op=MPI.SUM,root=0)
all_dets_table = None
if comm_rank == 0:
all_dets_table = Table()
all_dets_table.name = suffix
all_dets_table.cols = ['x', 'y', 'w', 'h', 'score', 'time', 'cls_ind', 'img_ind']
all_dets_table.arr = np.vstack(final_dets)
np.save(filename,all_dets_table)
print("Found %d dets"%all_dets_table.shape[0])
all_dets_table = comm.bcast(all_dets_table,root=0)
time_elapsed = time.time()-t
print("DatasetPolicy.load_ext_detections took %.3f"%time_elapsed)
return all_dets_table
def load_ext_detections(cls, dataset, suffix, force=False):
"""
Loads multi-image, multi-class array of detections for all images in the
given dataset.
Loads from canonical cache location.
"""
t = time.time()
filename = config.get_ext_dets_filename(dataset, suffix)
# check for cached full file
if os.path.exists(filename) and not force:
all_dets_table = np.load(filename)[()]
else:
# TODO also return times, or process them to add to dets?
all_dets = []
for i in range(comm_rank, len(dataset.images), comm_size):
image = dataset.images[i]
if re.search('dpm', suffix):
# NOTE: not actually using the given suffix in the call below
dets = cls.load_dpm_dets_for_image(image, dataset)
ind_vector = np.ones((np.shape(dets)[0], 1)) * i
dets = np.hstack((dets, ind_vector))
cols = [
'x', 'y', 'w', 'h', 'dummy', 'dummy', 'dummy', 'dummy',
'score', 'time', 'cls_ind', 'img_ind'
]
good_ind = [0, 1, 2, 3, 8, 9, 10, 11]
dets = dets[:, good_ind]
elif re.search('csc', suffix):
# NOTE: not actually using the given suffix in the call below
dets = cls.load_csc_dpm_dets_for_image(image, dataset)
ind_vector = np.ones((np.shape(dets)[0], 1)) * i
dets = np.hstack((dets, ind_vector))
elif re.search('ctf', suffix):
# Split the suffix into ctf and the main part
actual_suffix = suffix.split('_')[1]
dets = cls.load_ctf_dets_for_image(image, dataset,
actual_suffix)
ind_vector = np.ones((np.shape(dets)[0], 1)) * i
dets = np.hstack((dets, ind_vector))
else:
print(suffix)
raise RuntimeError('Unknown detector type in suffix')
all_dets.append(dets)
final_dets = None
if comm_rank == 0:
final_dets = []
safebarrier(comm)
final_dets = comm.reduce(all_dets, op=MPI.SUM, root=0)
all_dets_table = None
if comm_rank == 0:
all_dets_table = Table()
all_dets_table.name = suffix
all_dets_table.cols = [
'x', 'y', 'w', 'h', 'score', 'time', 'cls_ind', 'img_ind'
]
all_dets_table.arr = np.vstack(final_dets)
np.save(filename, all_dets_table)
print("Found %d dets" % all_dets_table.shape[0])
all_dets_table = comm.bcast(all_dets_table, root=0)
time_elapsed = time.time() - t
print("DatasetPolicy.load_ext_detections took %.3f" % time_elapsed)
return all_dets_table
from common_imports import * from common_mpi import * import synthetic.config as config from synthetic.dataset import Dataset from synthetic.classifier import Classifier if __name__=='__main__': train_set = 'full_pascal_train' train_dataset = Dataset(train_set) images = train_dataset.images classes = config.pascal_classes suffix = 'default' filename = config.get_ext_dets_filename(train_dataset, 'csc_'+suffix) csc_train = np.load(filename) csc_train = csc_train[()] csc_train = csc_train.subset(['score', 'cls_ind', 'img_ind']) score = csc_train.subset(['score']).arr classif = Classifier() csc_train.arr = classif.normalize_dpm_scores(csc_train.arr) numpos = train_dataset.get_ground_truth().shape[0] threshs = np.arange(0,1.01,0.05) result_filename = config.res_dir + 'thresh_classify.txt' result_file = open(result_filename, 'a')
from common_imports import *
from common_mpi import *
import synthetic.config as config
from synthetic.dataset import Dataset
from synthetic.classifier import Classifier
if __name__ == '__main__':
train_set = 'full_pascal_train'
train_dataset = Dataset(train_set)
images = train_dataset.images
classes = config.pascal_classes
suffix = 'default'
filename = config.get_ext_dets_filename(train_dataset, 'csc_' + suffix)
csc_train = np.load(filename)
csc_train = csc_train[()]
csc_train = csc_train.subset(['score', 'cls_ind', 'img_ind'])
score = csc_train.subset(['score']).arr
classif = Classifier()
csc_train.arr = classif.normalize_dpm_scores(csc_train.arr)
numpos = train_dataset.get_ground_truth().shape[0]
threshs = np.arange(0, 1.01, 0.05)
result_filename = config.res_dir + 'thresh_classify.txt'
result_file = open(result_filename, 'a')
threshs = np.array([0.15])