def snapshot(self):
output_dir = get_output_path(self.imdb, None)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
infix = ('_' + cfg.TRAIN.SNAPSHOT_INFIX
if cfg.TRAIN.SNAPSHOT_INFIX != '' else '')
filename = self.solver_param.snapshot_prefix + infix + \
'_iter_{:d}'.format(self.solver.iter) + '.caffemodel'
filename = os.path.join(output_dir, filename)
self.solver.net.save(str(filename))
print 'Wrote snapshot to: {:s}'.format(filename)
def test_net(net, imdb):
num_images = len(imdb.image_index)
output_dir = get_output_path(imdb, net)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
score_file = os.path.join(output_dir, 'fast_dbox_output_scores')
roidb = imdb.roidb
score_list = []
batch_size = 3000
# timers
_t = {'im_detect': Timer(), 'misc': Timer()}
sum_time = 0
for i in xrange(num_images):
im = cv2.imread(imdb.image_path_at(i))
num_gt = len(np.nonzero(roidb[i]['gt_classes'])[0])
boxes = roidb[i]['boxes'][num_gt:, :]
print 'Process image {:d},{:d} boxes'.format(i, boxes.shape[0])
_t['im_detect'].tic()
#Pass the boxes through the net in batches
if boxes.shape[0] % batch_size == 0:
rid = boxes.shape[0] / batch_size
else:
rid = boxes.shape[0] / batch_size + 1
for j in xrange(rid):
start_ind = batch_size * j
end_ind = min(start_ind + batch_size, boxes.shape[0])
boxes_ = boxes[start_ind:end_ind]
scores_ = im_obj_detect(net, im, boxes_)
if (j == 0):
scores = scores_
else:
scores = np.concatenate((scores, scores_), axis=0)
_t['im_detect'].toc()
print 'image {:d}/{:d} {:.3f}s'.format(i + 1, num_images,
_t['im_detect'].average_time)
sum_time = sum_time + _t['im_detect'].average_time
score_list.append(scores)
print 'Average time = {:.3f}s'.format(sum_time / num_images)
#Save Edge boxes score in both .mat and .pkl format
sio.savemat(score_file + '.mat', {'score_list': score_list})
f_score = open(score_file + '.pkl', 'wb')
cPickle.dump(score_list, f_score)
f_score.close()
def test_net(net, imdb):
num_images = len(imdb.image_index)
output_dir = get_output_path(imdb, net)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
score_file = os.path.join(output_dir, 'fast_dbox_output_scores')
roidb = imdb.roidb
score_list = []
batch_size = 3000
# timers
_t = {'im_detect' : Timer(), 'misc' : Timer()}
sum_time = 0
for i in xrange(num_images):
im = cv2.imread(imdb.image_path_at(i))
num_gt = len(np.nonzero(roidb[i]['gt_classes'])[0])
boxes = roidb[i]['boxes'][num_gt:,:]
print 'Process image {:d},{:d} boxes'.format(i,boxes.shape[0])
_t['im_detect'].tic()
#Pass the boxes through the net in batches
if boxes.shape[0] % batch_size == 0:
rid = boxes.shape[0]/batch_size
else:
rid = boxes.shape[0]/batch_size+1
for j in xrange(rid):
start_ind = batch_size*j
end_ind = min(start_ind+batch_size,boxes.shape[0])
boxes_ = boxes[start_ind:end_ind]
scores_ = im_obj_detect(net, im, boxes_)
if (j==0):
scores = scores_
else:
scores = np.concatenate((scores, scores_), axis=0)
_t['im_detect'].toc()
print 'image {:d}/{:d} {:.3f}s'.format(i+1,num_images,_t['im_detect'].average_time)
sum_time = sum_time+_t['im_detect'].average_time
score_list.append(scores)
print 'Average time = {:.3f}s'.format(sum_time/num_images)
#Save Edge boxes score in both .mat and .pkl format
sio.savemat(score_file+'.mat', {'score_list':score_list})
f_score = open(score_file+'.pkl','wb')
cPickle.dump(score_list,f_score)
f_score.close()
