def get_segmentations(self, img_id):
img_id_str = self.get_str_id(img_id)
run_name = img_id_str.split('_')[0]
gt_fname = os.path.join(
self.gt_folder, run_name,
img_id_str + '_{}_instanceIds.png'.format(self.gt_subfolder))
gt_img = cv2.imread(gt_fname, -1)
if gt_img is None:
log.warning('GT image does not exist: "{}"'.format(gt_fname))
# raise Exception('Image file not exist: "{}"'.format(gt_fname))
segm = []
colors = []
else:
# log.warning('GT image does exist: "{}"'.format(gt_fname))
segm, colors = sep_labels.get_separate_labels(gt_img)
sem_segm = [None] * 8
segm_final = []
segm_sem_cls = []
for ss, cc in zip(segm, colors):
if cc > 1000:
sem_cls = int(np.floor(cc / 1000))
ins_id = int(cc) % 1000
label = self.id_to_label[sem_cls]
if label.trainId > 0:
sem_cls_train = label.trainId
segm_final.append(ss)
if sem_segm[sem_cls_train - 1] is None:
sem_segm[sem_cls_train - 1] = np.zeros(ss.shape)
sem_segm[sem_cls_train - 1] = np.maximum(sem_segm[sem_cls_train - 1],
ss)
segm_sem_cls.append(sem_cls_train - 1)
return segm_final, sem_segm, segm_sem_cls
def get_segmentations(self, img_id):
img_id_str = self.get_str_id(img_id)
fname = '{}.png'.format(img_id_str)
gt_fname = os.path.join(self.gt_folder, fname)
if not os.path.exists(gt_fname):
raise Exception('GT file not exists: {} or '.format(gt_fname))
gt_img = cv2.imread(gt_fname)
segm, colors = sep_labels.get_separate_labels(gt_img)
if len(segm) > 0:
sem_segm = [np.zeros(segm[0].shape)]
for ss in segm:
sem_segm[0] = np.maximum(ss, sem_segm[0])
else:
sem_segm = []
return segm, sem_segm, [0] * len(segm)
def get_segmentations(self, img_id):
img_id_str = self.get_str_id(img_id)
run_name = img_id_str.split("_")[0]
if self.semantic_only:
gt_fname_suffix = "labelIds.png"
else:
gt_fname_suffix = "instanceIds.png"
gt_fname = os.path.join(
self.gt_folder, run_name,
img_id_str + "_{}_{}".format(self.gt_subfolder, gt_fname_suffix))
gt_img = cv2.imread(gt_fname, -1)
if gt_img is None:
log.warning("GT image does not exist: \"{}\"".format(gt_fname))
segm = []
colors = []
else:
segm, colors = sep_labels.get_separate_labels(gt_img)
sem_segm = {}
segm_final = []
segm_sem_cls = []
for ss, cc in zip(segm, colors):
if cc > 1000:
sem_cls = int(np.floor(cc / 1000))
ins_id = int(cc) % 1000
else:
sem_cls = cc
ins_id = 0
label = self.id_to_label[sem_cls]
if label.trainId >= 0:
sem_cls_train = label.trainId
segm_final.append(ss)
if sem_cls_train not in sem_segm:
sem_segm[sem_cls_train] = np.zeros(ss.shape)
sem_segm[sem_cls_train] = np.maximum(sem_segm[sem_cls_train], ss)
segm_sem_cls.append(sem_cls_train)
return segm_final, sem_segm, segm_sem_cls