import sys
sys.path.insert(0, "../utils/")
import numpy as np
import os
import glob
import scipy.io as sio
import cv2
from skimage import io
from utils import cv_crop
import torch
from joblib import Parallel, delayed
def transform(point, center, scale, resolution, rotation=0, invert=False):
_pt = np.ones(3)
_pt[0] = point[0]
_pt[1] = point[1]
h = 200.0 * scale
t = np.eye(3)
t[0, 0] = resolution / h
t[1, 1] = resolution / h
t[0, 2] = resolution * (-center[0] / h + 0.5)
t[1, 2] = resolution * (-center[1] / h + 0.5)
if rotation != 0:
rotation = -rotation
r = np.eye(3)
ang = rotation * math.pi / 180.0
s = math.sin(ang)
c = math.cos(ang)
r[0][0] = c
r[0][1] = -s
r[1][0] = s
r[1][1] = c
t_ = np.eye(3)
t_[0][2] = -resolution / 2.0
t_[1][2] = -resolution / 2.0
t_inv = torch.eye(3)
t_inv[0][2] = resolution / 2.0
t_inv[1][2] = resolution / 2.0
t = reduce(np.matmul, [t_inv, r, t_, t])
if invert:
t = np.linalg.inv(t)
new_point = (np.matmul(t, _pt))[0:2]
return new_point.astype(float)
def parse_pts(pts_file):
pts = []
with open(pts_file) as f:
for line in f.readlines():
line = line.strip()
if line[0].isdigit() == False:
continue
else:
idx = line.find(' ')
x, y = float(line[:idx]), float(line[idx+1:])
pts.append([x, y])
if len(pts) != 68:
print('Not enough points')
else:
return np.array(pts)
class WFLWInstance():
def __init__(self, line, idx):
self.idx = idx
line = line.strip().split(' ')
# convert landmarks
landmarks_list = list(map(float, line[:196]))
self.landmarks = []
for i in range(0, 196, 2):
self.landmarks.append([landmarks_list[i], landmarks_list[i+1]])
self.landmarks = np.array(self.landmarks)
# convert bboxes
if len(line) == 207:
self.bbox = list(map(float, line[196:200]))
else:
self.bbox = None
# convert image name
self.image_base_name = line[-1]
self.image_first_point = line[0]
def load_meta_subset_data(meta_path):
with open(meta_path) as f:
lines = f.readlines()
meta_data = []
idx = 0
for line in lines:
line = line.strip().split(' ')
meta_data.append(line[-1]+line[0])
return meta_data
def load_meta_data(meta_path, meta_subset_data=None):
with open(meta_path) as f:
lines = f.readlines()
meta_data = []
idx = 0
for line in lines:
wflw_instance = WFLWInstance(line, idx)
if meta_subset_data is not None and (wflw_instance.image_base_name+wflw_instance.image_first_point) in meta_subset_data:
meta_data.append(wflw_instance)
idx += 1
return meta_data
def process_single(single, image_path, image_save_path, landmarks_save_path):
# print('Processing: {}'.format(single.image_base_name))
image_full_path = os.path.join(image_path, single.image_base_name)
image = io.imread(image_full_path)
if len(image.shape) == 2:
image = np.stack((image, image, image), -1)
pts = single.landmarks
left, top, right, bottom = [int(x) for x in single.bbox]
lr_pad = int(0.05 * (right - left) / 2)
tb_pad = int(0.05 * (bottom - top) / 2)
left = max(0, left - lr_pad)
right = right + lr_pad
top = max(0, top - tb_pad)
bottom = bottom + tb_pad
center = torch.FloatTensor(
[right - (right - left) / 2.0, bottom -
(bottom - top) / 2.0])
scale_factor = 250.0
scale = (right - left + bottom - top) / scale_factor
new_image, new_landmarks = cv_crop(image, pts, center, scale, 450, 0)
while np.min(new_landmarks) < 10 or np.max(new_landmarks) > 440:
scale_factor -= 10
scale = (right - left + bottom - top) / scale_factor
new_image, new_landmarks = cv_crop(image, pts, center, scale, 450, 0)
assert (scale_factor > 0), "Landmarks out of boundary!"
if new_image != []:
io.imsave(os.path.join(image_save_path, os.path.basename(image_full_path[:-4]+'_' + str(single.idx) + image_full_path[-4:])), new_image)
np.save(os.path.join(landmarks_save_path, os.path.basename(image_full_path[:-4]+ '_' + str(single.idx) + '.pts')), new_landmarks)
if __name__ == '__main__':
image_path = './WFLW_images/'
meta_subset_path = './WFLW_annotations/list_98pt_rect_attr_train_test/list_98pt_rect_attr_test.txt'
meta_path = './WFLW_annotations/list_98pt_rect_attr_train_test/list_98pt_rect_attr_test.txt'
image_save_path = './WFLW_test/images/'
landmarks_save_path = './WFLW_test/landmarks/'
if not os.path.exists(image_save_path):
os.makedirs(image_save_path)
if not os.path.exists(landmarks_save_path):
os.makedirs(landmarks_save_path)
exts = ['*.png', '*.jpg']
meta_subset_data = load_meta_subset_data(meta_subset_path)
meta_data = load_meta_data(meta_path, meta_subset_data)
assert (len(meta_data) == len(meta_subset_data)), "Some images are missing!"
print("Total images: {0:d}".format(len(meta_data)))
Parallel(n_jobs=10,
backend='threading',
verbose=10)(delayed(process_single)(single, image_path,
image_save_path,
landmarks_save_path) for single in meta_data)
