def generate_batch_sample(input_dir, save_folder):
uv_h = uv_w = 256
# load uv coords
global uv_coords
uv_coords = face3d.morphable_model.load.load_uv_coords(
'./examples/Data/BFM/Out/BFM_UV.mat') #
uv_coords = process_uv(uv_coords, uv_h, uv_w)
# load bfm
bfm = MorphabelModel('./examples/Data/BFM/Out/BFM.mat')
base = 0
path_w = open(
'/media/weepies/Seagate Backup Plus Drive/3DMM/train_path_afw.txt',
'w')
print('input: ', input_dir)
for idx, item in enumerate(os.listdir(input_dir)):
print('dealing')
if 'jpg' in item:
ab_path = os.path.join(input_dir, item)
img_path = ab_path
mat_path = ab_path.replace('jpg', 'mat')
run_posmap_300W_LP(bfm, img_path, mat_path, save_folder,
idx + base)
name_p = image_name = img_path.strip().split('/')[-1]
num_name = name_p.split('.jpg')
path_w.write(
'/media/weepies/Seagate Backup Plus Drive/3DMM/posnet/300W_HELEN/'
+ name_p + '*' +
'/media/weepies/Seagate Backup Plus Drive/3DMM/posnet/300W_HELEN/'
+ num_name[0] + '.txt' + '\n')
# print("Number {} uv_pos_map was generated!".format(idx))
path_w.close()
def make_posmaps(num_to_read=10, folder='300W_LP'):
uv_h = uv_w = 256
image_h = image_w = 256
save_folder = 'data_input'
if folder == 'AFLW2000':
save_folder = 'posmap_output'
global uv_coords
uv_coords = face3d.morphable_model.load.load_uv_coords(
'BFM/Out/BFM_UV.mat')
uv_coords = process_uv(uv_coords, uv_h, uv_w)
bfm = MorphabelModel('BFM/Out/BFM.mat')
img_paths = glob.glob(folder + '/*.jpg')
mat_paths = glob.glob(folder + '/*.mat')
img_mat_comb = []
for i in range(len(img_paths)):
img_paths[i] = img_paths[i].replace('\\', '/')
mat_paths[i] = mat_paths[i].replace('\\', '/')
img_mat_comb.append([img_paths[i], mat_paths[i]])
random.shuffle(img_mat_comb)
for i in range(num_to_read):
run_posmap_300W_LP(bfm, img_mat_comb[i][0], img_mat_comb[i][1],
save_folder)
def generate_batch_sample(input_dir, save_folder='./300WLP'):
if not os.path.exists(save_folder):
os.mkdir(save_folder)
# set para
uv_h = uv_w = 256
# load uv coords
global uv_coords
uv_coords = face3d.morphable_model.load.load_uv_coords('BFM/BFM_UV.mat') #
uv_coords = process_uv(uv_coords, uv_h, uv_w)
# load bfm
bfm = MorphabelModel('BFM/BFM.mat')
# Batch generating uv_map Dataset
"""
@date: 2019/07/19
Train Dataset:
AFW. 10413.
HELEN. 75351.
LFPW. 33111.
Test Dataset:
IBUG. 3571.
"""
base = 0
for idx, item in enumerate(os.listdir(input_dir)):
if 'jpg' in item:
ab_path = os.path.join(input_dir, item)
img_path = ab_path
mat_path = ab_path.replace('jpg', 'mat')
run_posmap_300W_LP(bfm, img_path, mat_path, save_folder, idx + base)
print("Number {} uv_pos_map was generated!".format(idx))
def get_depth_image(image_path):
im = cv2.imread(image_path)
h, w, c = im.shape
landmarks = pd.read_pickle(image_path.replace('.jpg', '.pkl'))
bfm = MorphabelModel('Data/BFM/Out/BFM.mat')
x = mesh.transform.from_image(landmarks, h, w)
X_ind = bfm.kpt_ind
fitted_sp, fitted_ep, fitted_s, fitted_angles, fitted_t = bfm.fit(
x, X_ind, max_iter=200, isShow=False)
colors = bfm.generate_colors(np.random.rand(bfm.n_tex_para, 1))
colors = np.minimum(np.maximum(colors, 0), 1)
fitted_vertices = bfm.generate_vertices(fitted_sp, fitted_ep)
transformed_vertices = bfm.transform(fitted_vertices, fitted_s,
fitted_angles, fitted_t)
image_vertices = mesh.transform.to_image(transformed_vertices, h, w)
triangles = bfm.triangles
z = image_vertices[:, 2:]
z = z - np.min(z)
z = z / np.max(z)
attribute = z
depth_image = mesh.render.render_colors(image_vertices,
triangles,
attribute,
h,
w,
c=1)
depth_image = (depth_image * 255).astype('uint8')
savename = image_path.replace('.jpg', '-depth.jpg')
io.imsave(savename, np.squeeze(depth_image))
def generate_bfm_base_obj():
bfm = MorphabelModel('../Data/BFM/Out/BFM.mat')
sp = bfm.get_shape_para('zero')
ep = bfm.get_exp_para('zero')
tp = bfm.get_tex_para('random')
vertices = bfm.generate_vertices(sp, ep)
colors = bfm.generate_colors(tp)
colors = np.minimum(np.maximum(colors, 0), 255)
write_obj_with_colors('head_fit.obj', vertices, bfm.triangles, colors)
def generate_prnet_trainset(root_300wlp, save_trainset):
'''
Input：
root_300wlp. The root path of your 300w_lp.
save_trainset. The path you want to save.
(All train set totally about 80G)
'''
root_300wlp = os.path.abspath(root_300wlp).replace('\\',
'/') # For windows
save_trainset = os.path.abspath(save_trainset).replace('\\',
'/') # For windows
# set para
uv_h = uv_w = 256
# load uv coords
global uv_coords
uv_coords = face3d.morphable_model.load.load_uv_coords(
'Data/BFM/Out/BFM_UV.mat')
uv_coords = process_uv(uv_coords, uv_h, uv_w)
# load bfm
bfm = MorphabelModel('Data/BFM/Out/BFM.mat')
# run
fp_label = open('trainDataLabel.txt', "w")
sub_dir_list = os.listdir(root_300wlp)
for item in sub_dir_list:
save_folder = os.path.join(save_trainset, item)
if not os.path.exists(save_folder):
os.makedirs(save_folder)
img_list_rep = os.path.join(root_300wlp, item, "*.jpg")
img_list = glob.glob(img_list_rep)
for img_path in img_list:
img_path = img_path.replace('\\', '/') # For windows
mat_path = img_path.replace('jpg', 'mat')
if not os.path.exists(mat_path):
continue
save_img_path = img_path.replace(root_300wlp, save_trainset)
save_npy_path = save_img_path.replace('jpg', 'npy')
if os.path.exists(save_img_path) and os.path.exists(save_npy_path):
fp_label.writelines(save_img_path + ' ' + save_npy_path + '\n')
print("Passing ...")
continue
run_posmap_300W_LP(bfm, img_path, mat_path, save_folder)
print(save_img_path)
fp_label.writelines(save_img_path + ' ' + save_npy_path + '\n')
fp_label.close()
def demo():
save_folder = 'results/posmap_300WLP'
if not os.path.exists(save_folder):
os.mkdir(save_folder)
# set para
uv_h = uv_w = 256
# load uv coords
global uv_coords
uv_coords = face3d.morphable_model.load.load_uv_coords('Data/BFM/Out/BFM_UV.mat')
uv_coords = process_uv(uv_coords, uv_h, uv_w)
# load bfm
bfm = MorphabelModel('Data/BFM/Out/BFM.mat')
# run
image_path = 'Data/IBUG_image_008_1_0.jpg'
mat_path = 'Data/IBUG_image_008_1_0.mat'
run_posmap_300W_LP(bfm, image_path, mat_path, save_folder)
def __init__(self, model_dir, render_only=False):
self.bfm = MorphabelModel(osp.join(model_dir, 'BFM.mat'))
self.index_ind = self.bfm.kpt_ind
uv_coords = face3d.morphable_model.load.load_uv_coords(
osp.join(model_dir, 'BFM_UV.mat'))
self.uv_size = (224, 224)
self.mask_stxr = 0.1
self.mask_styr = 0.33
self.mask_etxr = 0.9
self.mask_etyr = 0.7
self.tex_h, self.tex_w, self.tex_c = self.uv_size[1], self.uv_size[
0], 3
texcoord = np.zeros_like(uv_coords)
texcoord[:, 0] = uv_coords[:, 0] * (self.tex_h - 1)
texcoord[:, 1] = uv_coords[:, 1] * (self.tex_w - 1)
texcoord[:, 1] = self.tex_w - texcoord[:, 1] - 1
self.texcoord = np.hstack((texcoord, np.zeros((texcoord.shape[0], 1))))
self.X_ind = self.bfm.kpt_ind
if not render_only:
from image_3d68 import Handler
self.if3d68_handler = Handler(osp.join(model_dir, 'if1k3d68'),
0,
192,
ctx_id=0)
if __name__ == '__main__':
if not os.path.exists(FLAGS["save_uv_path"]):
os.mkdir(FLAGS["save_uv_path"])
if not os.path.exists(FLAGS["save_im_path"]):
os.mkdir(FLAGS["save_im_path"])
# load uv coords
global uv_coords
uv_coords = face3d.morphable_model.load.load_uv_coords(
FLAGS["bfm_uv_path"])
uv_coords = process_uv(uv_coords, FLAGS["uv_h"], FLAGS["uv_w"])
# load bfm
bfm = MorphabelModel(FLAGS["bfm_path"])
img_names_list = Path(
FLAGS["list_path"]).read_text().strip().split('\n')[:128]
if FLAGS["is62Param"]:
param_62d = pickle.load(open(FLAGS["param_path"], 'rb'))
index = 0
for img_name in tqdm(img_names_list):
file_name = os.path.splitext(img_name)[0]
image_path = os.path.join(FLAGS["data_path"], file_name + ".jpg")
param = param_62d[index]
generate_posmap_lb_62params(bfm, image_path, param,
FLAGS["save_uv_path"],
FLAGS["save_im_path"])
index = index + 1
else:
for img_name in tqdm(img_names_list):
import os, sys
import subprocess
import numpy as np
import scipy.io as sio
from skimage import io
from time import time
import matplotlib.pyplot as plt
sys.path.append('..')
import face3d
from face3d import mesh
from face3d.morphable_model import MorphabelModel
# --------------------- Forward: parameters(shape, expression, pose) --> 3D obj --> 2D image ---------------
# --- 1. load model
bfm = MorphabelModel('Data/BFM/Out/BFM.mat')
print('init bfm model success')
# --- 2. generate face mesh: vertices(represent shape) & colors(represent texture)
sp = bfm.get_shape_para('random')
ep = bfm.get_exp_para('random')
vertices = bfm.generate_vertices(sp, ep)
tp = bfm.get_tex_para('random')
colors = bfm.generate_colors(tp)
colors = np.minimum(np.maximum(colors, 0), 1)
# --- 3. transform vertices to proper position
s = 8e-04
angles = [10, 30, 20]
t = [0, 0, 0]
save_folder = '300W_LP_out'
if not os.path.exists(save_folder):
os.mkdir(save_folder)
# set para
uv_h = uv_w = 450
image_h = image_w = 450
# load uv coords
global uv_coords
uv_coords = face3d.morphable_model.load.load_uv_coords(
'./BFM/BFM_UV.mat') #
uv_coords = process_uv(uv_coords, uv_h, uv_w)
# load bfm
bfm = MorphabelModel('./BFM/BFM.mat')
# run
folders = [
"AFW", "AFW_Flip", "HELEN", "HELEN_Flip", "IBUG", "IBUG_Flip", "LFPW",
"LFPW_Flip"
]
folders = ["AFW", "IBUG"]
for i in folders:
folder_path = os.path.join(".", "300W_LP", i)
files = [j[:-4] for j in os.listdir(folder_path) if j[-4:] == ".jpg"]
for k in tqdm(files):
image_path = os.path.join(folder_path, k + ".jpg")
mat_path = os.path.join(folder_path, k + ".mat")
save_path = os.path.join(save_folder, i + "_" + k)
if not os.path.exists(save_path):
import dlib
import glob
import subprocess
import numpy as np
import scipy.io as sio
from skimage import io
from time import time
import matplotlib.pyplot as plt
sys.path.append('..')
import face3d
from face3d import mesh
from face3d.morphable_model import MorphabelModel
# load BFM model
bfm = MorphabelModel('examples/Data/BFM/Out/BFM.mat')
print('init bfm model success')
uv_coords = face3d.morphable_model.load.load_uv_coords(
'examples/Data/BFM/Out/BFM_UV.mat')
t = [0, 0, 0]
c = 3
#random texture and colors
tp = bfm.get_tex_para('random')
colors = bfm.generate_colors(tp)
colors = np.minimum(np.maximum(colors, 0), 1)
predictor_path = "shape_predictor_68_face_landmarks.dat"
faces_folder_path = sys.argv[1]
result_folder_path = sys.argv[2]
def main(args):
with open(args.img_list) as f:
img_list = [x.strip() for x in f.readlines()]
landmark_list = []
if not os.path.exists(args.save_dir):
os.mkdir(args.save_dir)
if not os.path.exists(args.save_lmk_dir):
os.mkdir(args.save_lmk_dir)
for img_idx, img_fp in enumerate(tqdm(img_list)):
im = cv2.imread(os.path.join(args.img_prefix, img_fp), 1)
gray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
h, w, c = im.shape
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor(
"face3d/models/shape_predictor_68_face_landmarks.dat")
rects = detector(gray, 1)
shape = predictor(gray, rects[0])
tl_corner_x = rects[0].center().x - rects[0].width() / 2
tl_corner_y = rects[0].center().y - rects[0].height() / 2
br_corner_x = rects[0].center().x + rects[0].width() / 2
br_corner_y = rects[0].center().y + rects[0].height() / 2
rects = [(tl_corner_x, tl_corner_y), (br_corner_x, br_corner_y)]
landmarks = np.zeros((68, 2))
for i, p in enumerate(shape.parts()):
landmarks[i] = [p.x, p.y]
im = cv2.circle(im, (p.x, p.y),
radius=3,
color=(0, 0, 255),
thickness=5)
bfm = MorphabelModel('face3d/Data/BFM/Out/BFM.mat')
x = mesh_numpy.transform.from_image(landmarks, h, w)
X_ind = bfm.kpt_ind
fitted_sp, fitted_ep, fitted_s, fitted_angles, fitted_t = bfm.fit(
x, X_ind, max_iter=200, isShow=False)
colors = bfm.generate_colors(np.random.rand(bfm.n_tex_para, 1))
colors = np.minimum(np.maximum(colors, 0), 1)
fitted_vertices = bfm.generate_vertices(fitted_sp, fitted_ep)
transformed_vertices = bfm.transform(fitted_vertices, fitted_s,
fitted_angles, fitted_t)
image_vertices = mesh_numpy.transform.to_image(transformed_vertices, h,
w)
triangles = bfm.triangles
colors = colors / np.max(colors)
attribute = colors
color_image = mesh_numpy.render.render_colors(image_vertices,
triangles,
attribute,
h,
w,
c=3)
io.imsave(os.path.join(args.save_lmk_dir,
str(img_idx) + ".png"),
img_as_ubyte(color_image))