def mean_scale(src,dst, mean_limb_file): fns, labels = read_labels(src) labels = np.array(labels) labels = np.reshape(labels,[labels.shape[0], labels.shape[1]/3, 3]) labels = mean_limb_scale(labels, mean_limb_file) labels = np.reshape(labels,[labels.shape[0], labels.shape[1] * labels.shape[2]]) with open(dst,'w') as dst_f: for i in range(len(fns)): write_str = labels[i] write_str = [str(x) for x in write_str] write_str = ','.join(write_str) write_str = fns[i] +' ' +write_str dst_f.write(write_str + '\n')
def normalize_label(gt_fn, dst_gt_fn, valid_data=False):
max_value, min_value = get_max_min()
fns = []
dst_file = open(dst_gt_fn, 'w')
sample_id = 0
fns, gt_labels = read_labels(gt_fn)
for i in range(len(fns)):
if valid_data and np.random.uniform() > valid_ratio:
continue
if valid_data and sample_id >= num_valid:
break
fn = fns[i]
labels = gt_labels[i]
labels = np.array(labels)
labels = (labels - min_value) / (max_value - min_value)
labels = labels.tolist()
labels = [str(x) for x in labels]
label_str = ','.join(labels)
dst_file.write(fn + ' ' + label_str + '\n')
sample_id += 1
def cal_mean_limb(src, dst):
parent_node = {
0: 0,
1: 0,
2: 1,
3: 2,
4: 0,
5: 4,
6: 5,
7: 0,
8: 7,
9: 8,
10: 9,
11: 8,
12: 11,
13: 12,
14: 8,
15: 14,
16: 15
}
pose_indexes = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]
fns, labels = read_labels(src)
labels = np.array(labels)
labels = np.reshape(labels, [labels.shape[0], labels.shape[1] / 3, 3])
limb_length = np.zeros((labels.shape[0], labels.shape[1]))
for i in range(labels.shape[0]):
for j in range(labels.shape[1]):
parent_id = parent_node[j]
limb_length[i][j] = np.linalg.norm(labels[i][j] -
labels[i][parent_id])
mean_limb_length = np.mean(limb_length, axis=0)
print(mean_limb_length)
with open(dst, 'w') as f:
mean_limb_length = mean_limb_length.tolist()
mean_limb_length = [str(x) for x in mean_limb_length]
line = ','.join(mean_limb_length)
f.write(line + '\n')
print(line)