def read_labels(self):
depth_label = cv2.imread(self.depth_gt[0], cv2.IMREAD_GRAYSCALE)
# leggi file di testo
with open(self.obstacles_gt[0], 'r') as f:
obstacles = f.readlines()
obstacles = [x.strip() for x in obstacles]
obstacles_label = np.zeros(shape=(5, 8, 10))
for obs in obstacles:
parsed_str_obs = obs.split(" ")
parsed_obs = np.zeros(shape=9)
i = 0
for n in parsed_str_obs:
if i < 2:
parsed_obs[i] = int(n)
elif i == 8:
parsed_obs[i] = Classes.str_to_class_enum(n)
else:
parsed_obs[i] = float(n)
i += 1
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
0] = 1.0 if parsed_obs[8] == 3 else 0.0 # class 3
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
1] = 1.0 if parsed_obs[8] == 4 else 0.0 # class 4
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
2] = 1.0 if parsed_obs[8] == 1 else 0.0 # class 1
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
3] = 1.0 if parsed_obs[8] == 2 else 0.0 # class 2
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 4] = parsed_obs[2] # x
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 5] = parsed_obs[3] # y
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 6] = parsed_obs[4] # w
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 7] = parsed_obs[5] # h
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 8] = parsed_obs[6] * 0.1 # m
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 9] = parsed_obs[7] * 0.1 # v
labels = {
"depth": np.expand_dims(depth_label, 2),
"obstacles": np.reshape(obstacles_label, (40, 10))
}
return labels
def labels_from_file_multiclass_4(obstacles_gt):
with open(obstacles_gt, 'r') as f:
obstacles = f.readlines()
obstacles = [x.strip() for x in obstacles]
obstacles_label = np.zeros(shape=(5, 8, 10))
for obs_ in obstacles:
parsed_str_obs = obs_.split(" ")
parsed_obs = np.zeros(shape=9)
i_ = 0
for n in parsed_str_obs:
if i_ < 2:
parsed_obs[i_] = int(n)
elif i_ == 8:
parsed_obs[i_] = Classes.str_to_class_enum(n)
else:
parsed_obs[i_] = float(n)
i_ += 1
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
0] = 1.0 if parsed_obs[8] == 3 else 0.0 # class 3
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
1] = 1.0 if parsed_obs[8] == 4 else 0.0 # class 4
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
2] = 1.0 if parsed_obs[8] == 1 else 0.0 # class 1
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
3] = 1.0 if parsed_obs[8] == 2 else 0.0 # class 2
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 4] = parsed_obs[2] # x
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 5] = parsed_obs[3] # y
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 6] = parsed_obs[4] # w
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 7] = parsed_obs[5] # h
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 8] = parsed_obs[6] * 0.1 # m
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 9] = parsed_obs[7] * 0.1 # v
labels = np.reshape(obstacles_label, (40, 10))
return labels
def labels_from_file_multiclass_2(obstacles_gt):
with open(obstacles_gt, 'r') as f:
obstacles = f.readlines()
obstacles = [x.strip() for x in obstacles]
obstacles_label = np.zeros(shape=(5, 8, 8))
for obs_ in obstacles:
parsed_str_obs = obs_.split(" ")
parsed_obs = np.zeros(shape=9)
i_ = 0
for n in parsed_str_obs:
if i_ < 2:
parsed_obs[i_] = int(n)
elif i_ == 8:
if n == 'robot_team' or n == 'robot_opponent':
n = 'robot'
elif n == 'goal':
n = 'nothing'
parsed_obs[i_] = Classes.str_to_class_enum(n)
else:
parsed_obs[i_] = float(n)
i_ += 1
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
0] = 1.0 if parsed_obs[8] == 0 else 0.0 # class 0
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]),
1] = 1.0 if parsed_obs[8] == 1 else 0.0 # class 1
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 2] = parsed_obs[2] # x
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 3] = parsed_obs[3] # y
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 4] = parsed_obs[4] # w
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 5] = parsed_obs[5] # h
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 6] = parsed_obs[6] * 0.1 # m
obstacles_label[int(parsed_obs[1]),
int(parsed_obs[0]), 7] = parsed_obs[7] * 0.1 # v
labels = np.reshape(obstacles_label, (40, 8))
return labels
def read_labels_gt_viewer_multiclass(obstacles_gt, number_classes):
with open(obstacles_gt, 'r') as f:
obstacles = f.readlines()
obstacles = [x.strip() for x in obstacles]
labels = []
for obs in obstacles:
parsed_str_obs = obs.split(" ")
parsed_obs = np.zeros(shape=9)
i = 0
for n in parsed_str_obs:
if i < 2:
parsed_obs[i] = int(n)
elif i == 8:
if (number_classes == 2 or number_classes
== 3) and (n == 'robot_team' or n == 'robot_opponent'):
n = 'robot'
elif number_classes == 2 and n == 'goal':
n = 'nothing'
parsed_obs[i] = Classes.str_to_class_enum(n)
else:
parsed_obs[i] = float(n)
i += 1
x = int(parsed_obs[0] * 32 + parsed_obs[2] * 32)
y = int(parsed_obs[1] * 32 + parsed_obs[3] * 32)
w = int(parsed_obs[4] * 256)
h = int(parsed_obs[5] * 160)
object = [[x - w / 2, y - h / 2, w, h], [parsed_obs[6], parsed_obs[7]],
parsed_obs[8]]
# Object with last value equal to -1 is a dispensable object
if object[-1] != -1:
labels.append(object)
return labels
def read_labels_gt_viewer_multiclass_2(obstacles_gt):
with open(obstacles_gt, 'r') as f:
obstacles = f.readlines()
obstacles = [x.strip() for x in obstacles]
labels = []
for obs in obstacles:
parsed_str_obs = obs.split(" ")
parsed_obs = np.zeros(shape=9)
i = 0
for n in parsed_str_obs:
if i < 2:
parsed_obs[i] = int(n)
elif i == 8:
if n == 'robot_team' or n == 'robot_opponent':
n = 'robot'
parsed_obs[i] = Classes.str_to_class_enum(n)
else:
parsed_obs[i] = float(n)
i += 1
if parsed_obs[8] == 2:
continue
x = int(parsed_obs[0] * 32 + parsed_obs[2] * 32)
y = int(parsed_obs[1] * 32 + parsed_obs[3] * 32)
w = int(parsed_obs[4] * 256)
h = int(parsed_obs[5] * 160)
object = [[x - w / 2, y - h / 2, w, h], [parsed_obs[6], parsed_obs[7]],
parsed_obs[8]]
labels.append(object)
return labels
depth_gt = gt.copy()
#evaluate only on valid predictions (some methods like Cadena's may return zero or negative values)
depth_gt[np.nonzero(pred_depth_ <= 0)] = 0.0
else:
depth_gt = None
#Corrected depth
if results[2] is not None:
corr_depth = results[2][0, :, :, 0].copy()
corr_depth[np.nonzero(depth_gt <= 0)] = 0.0
results[2] = corr_depth
#Get obstacles from GT segmentation and depth
#gt_obs = EvaluationUtils.get_obstacles_from_seg_and_depth(gt, seg, segm_thr=-1)
if model_name == 'odl':
for ob in obs:
ob[2] = Classes.generate_class(ob[2])
gt_obs = EvaluationUtils.get_obstacles_from_list_multiclass(obs)
else:
gt_obs = EvaluationUtils.get_obstacles_from_list(obs)
if confMatrix:
obs_labels = []
if number_classes == 2:
obs_labels = labels_from_file_multiclass_2(obs_path)
elif number_classes == 3:
obs_labels = labels_from_file_multiclass_3(obs_path)
elif number_classes == 4:
obs_labels = labels_from_file_multiclass_4(obs_path)
conf_list_pred = []