torchnet / extract_lip.py
milselarch's picture
push to main
df07554
raw
history blame
4.79 kB
import cv2
import json
import numpy as np
from multiprocessing import Pool, Process, Queue
import time
import os
def get_position(size, padding=0.25):
x = [0.000213256, 0.0752622, 0.18113, 0.29077, 0.393397, 0.586856, 0.689483, 0.799124,
0.904991, 0.98004, 0.490127, 0.490127, 0.490127, 0.490127, 0.36688, 0.426036,
0.490127, 0.554217, 0.613373, 0.121737, 0.187122, 0.265825, 0.334606, 0.260918,
0.182743, 0.645647, 0.714428, 0.793132, 0.858516, 0.79751, 0.719335, 0.254149,
0.340985, 0.428858, 0.490127, 0.551395, 0.639268, 0.726104, 0.642159, 0.556721,
0.490127, 0.423532, 0.338094, 0.290379, 0.428096, 0.490127, 0.552157, 0.689874,
0.553364, 0.490127, 0.42689]
y = [0.106454, 0.038915, 0.0187482, 0.0344891, 0.0773906, 0.0773906, 0.0344891,
0.0187482, 0.038915, 0.106454, 0.203352, 0.307009, 0.409805, 0.515625, 0.587326,
0.609345, 0.628106, 0.609345, 0.587326, 0.216423, 0.178758, 0.179852, 0.231733,
0.245099, 0.244077, 0.231733, 0.179852, 0.178758, 0.216423, 0.244077, 0.245099,
0.780233, 0.745405, 0.727388, 0.742578, 0.727388, 0.745405, 0.780233, 0.864805,
0.902192, 0.909281, 0.902192, 0.864805, 0.784792, 0.778746, 0.785343, 0.778746,
0.784792, 0.824182, 0.831803, 0.824182]
x, y = np.array(x), np.array(y)
x = (x + padding) / (2 * padding + 1)
y = (y + padding) / (2 * padding + 1)
x = x * size
y = y * size
return np.array(list(zip(x, y)))
def cal_area(anno):
return (
(anno[:, 0].max() - anno[:, 0].min()) *
(anno[:, 1].max() - anno[:, 1].min())
)
def transformation_from_points(points1, points2):
points1 = points1.astype(np.float64)
points2 = points2.astype(np.float64)
c1 = np.mean(points1, axis=0)
c2 = np.mean(points2, axis=0)
points1 -= c1
points2 -= c2
s1 = np.std(points1)
s2 = np.std(points2)
points1 /= s1
points2 /= s2
U, S, Vt = np.linalg.svd(points1.T * points2)
R = (U * Vt).T
return np.vstack([np.hstack((
(s2 / s1) * R, c2.T - (s2 / s1) * R * c1.T)),
np.matrix([0., 0., 1.])
])
def anno_img(img_dir, anno_dir, save_dir):
files = list(os.listdir(img_dir))
files = [file for file in files if (file.find('.jpg') != -1)]
shapes = []
for file in files:
img = os.path.join(img_dir, file)
anno = os.path.join(anno_dir, file).replace('.jpg', '.txt')
I = cv2.imread(img)
count = 0
with open(anno, 'r') as f:
annos = [line.strip().split('\t') for line in f.readlines()]
if len(annos) == 0: return
for (i, anno) in enumerate(annos):
x, y = [], []
for p in anno:
_, __ = p[1:-1].split(',')
_, __ = float(_), float(__)
x.append(_)
y.append(__)
annos[i] = np.stack([x, y], 1)
anno = sorted(annos, key=cal_area, reverse=True)[0]
shape = []
shapes.append(anno[17:])
front256 = get_position(256)
M_prev = None
for (shape, file) in zip(shapes, files):
img = os.path.join(img_dir, file)
I = cv2.imread(img)
M = transformation_from_points(np.matrix(shape), np.matrix(front256))
img = cv2.warpAffine(I, M[:2], (256, 256))
(x, y) = front256[-20:].mean(0).astype(np.int32)
w = 160 // 2
img = img[y - w // 2:y + w // 2, x - w:x + w, ...]
cv2.imwrite(os.path.join(save_dir, file), img)
def run(files):
tic = time.time()
count = 0
print('n_files:{}'.format(len(files)))
for (img_dir, anno_dir, save_dir) in files:
anno_img(img_dir, anno_dir, save_dir)
count += 1
if count % 1000 == 0:
print('eta={}'.format(
(time.time() - tic) /
(count) * (len(files) - count) /
3600.0
))
if __name__ == '__main__':
with open('grid.txt', 'r') as f:
data = [line.strip() for line in f.readlines()]
data = list(set([os.path.split(file)[0] for file in data]))
annos = [name.replace('GRID/6k_video_imgs', 'GRID/landmarks') for name in data]
targets = [name.replace('GRID/6k_video_imgs', 'GRID/lip') for name in data]
for dst in targets:
if (not os.path.exists(dst)):
os.makedirs(dst)
data = list(zip(data, annos, targets))
processes = []
n_p = 8
bs = len(data) // n_p
for i in range(n_p):
if i == n_p - 1:
bs = len(data)
p = Process(target=run, args=(data[:bs],))
data = data[bs:]
p.start()
processes.append(p)
assert (len(data) == 0)
for p in processes:
p.join()