Spaces:
Running
Running
File size: 8,415 Bytes
44925e5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 |
import numpy as np
import tqdm
import os, glob
import json
import argparse
from render_utils.lib.networks.smpl_torch import SmplTorch
from render_utils.lib.utils.gaussian_np_utils import load_gaussians_from_ply
from render_utils.stitch_body_and_head import load_body_params, load_face_params, get_smpl_verts_and_head_transformation, calc_livehead2livebody
def load_rendering_camera(camera_fpath):
with open(camera_fpath, 'r') as fp:
camera_data = json.load(fp)
camera_data = camera_data[0]
image_size = [camera_data['width'], camera_data['height']]
cam_f = [camera_data['fx'], camera_data['fy']]
cam_pos = np.array(camera_data['position'])
cam_rot = np.array(camera_data['rotation']).reshape(3, 3)
c2w = np.eye(4)
c2w[:3, :3] = cam_rot
c2w[:3, 3] = cam_pos
cam_extr = np.linalg.inv(c2w)
cam_intr = np.eye(3)
cam_intr[0, 0] = cam_f[0]
cam_intr[1, 1] = cam_f[1]
cam_intr[0, 2] = image_size[0] / 2
cam_intr[1, 2] = image_size[1] / 2
return cam_extr, cam_intr, image_size
def load_camera_list(camera_fpath):
with open(camera_fpath, 'r') as fp:
camera_data = json.load(fp)
image_size = [camera_data[0]['width'], camera_data[0]['height']]
cam_list = []
for cam in camera_data:
cam_f = [cam['fx'], cam['fy']]
cam_pos = np.array(cam['position'])
cam_rot = np.array(cam['rotation']).reshape(3, 3)
c2w = np.eye(4)
c2w[:3, :3] = cam_rot
c2w[:3, 3] = cam_pos
cam_extr = np.linalg.inv(c2w)
cam_intr = np.eye(3)
cam_intr[0, 0] = cam_f[0]
cam_intr[1, 1] = cam_f[1]
cam_intr[0, 2] = image_size[0] / 2
cam_intr[1, 2] = image_size[1] / 2
cam_list.append((cam_extr, cam_intr))
return cam_list, image_size
def load_camera_data(cam):
image_size = [cam['width'], cam['height']]
cam_f = [cam['fx'], cam['fy']]
cam_pos = np.array(cam['position'])
cam_rot = np.array(cam['rotation']).reshape(3, 3)
c2w = np.eye(4)
c2w[:3, :3] = cam_rot
c2w[:3, 3] = cam_pos
cam_extr = np.linalg.inv(c2w)
cam_intr = np.eye(3)
cam_intr[0, 0] = cam_f[0]
cam_intr[1, 1] = cam_f[1]
cam_intr[0, 2] = image_size[0] / 2
cam_intr[1, 2] = image_size[1] / 2
return (cam_extr, cam_intr), image_size
def calc_offline_rendering_param(
body_gaussian_root_dir, ref_head_gaussian_path, ref_head_param_path, render_cam_fpath,
body_head_blending_param_path):
body_param_flist = sorted(glob.glob(os.path.join(body_gaussian_root_dir, 'posed_params/*.npz')))
head_gaussians = load_gaussians_from_ply(ref_head_gaussian_path)
head_pose, head_scale, id_coeff, exp_coeff = load_face_params(ref_head_param_path)
# cam_extr_body, cam_intr_body, image_size = load_rendering_camera(render_cam_fpath)
cam_list, image_size = load_camera_list(render_cam_fpath)
body_head_blending_params = np.load(body_head_blending_param_path)
smplx_to_faceverse = body_head_blending_params['smplx_to_faceverse']
residual_transf = body_head_blending_params['residual_transf']
body_nonface_mask = body_head_blending_params['body_nonface_mask']
head_nonface_mask = body_head_blending_params['head_nonface_mask']
head_facial_idx = body_head_blending_params['head_facial_idx']
body_facial_idx = body_head_blending_params['body_facial_idx']
head_body_corr_idx = body_head_blending_params['head_body_corr_idx']
head_color_bw = body_head_blending_params['head_color_bw']
color_transfer = body_head_blending_params['color_transfer']
smpl = SmplTorch(model_file='./AnimatableGaussians/smpl_files/smplx/SMPLX_NEUTRAL.npz')
head_cam_extr = []
head_cam_intr = []
head_cam_intr_zoom = []
head_zoom_center = []
head_zoom_scale = []
for i, body_param_fpath in enumerate(tqdm.tqdm(body_param_flist)):
global_orient, transl, body_pose, betas = load_body_params(body_param_fpath)
# body_gaussians = load_gaussians_from_ply(body_gaussian_fpath)
smpl_verts, head_joint_transfmat = get_smpl_verts_and_head_transformation(
smpl, global_orient, body_pose, transl, betas)
livehead2livebody = calc_livehead2livebody(head_pose, smplx_to_faceverse, head_joint_transfmat)
total_transf = np.matmul(livehead2livebody, residual_transf)
cam_extr = np.matmul(cam_list[i][0], total_transf)
cam_intr = np.copy(cam_list[i][1])
head_cam_extr.append(cam_extr)
head_cam_intr.append(cam_intr)
pts = np.copy(head_gaussians.xyz)
pts_proj = np.matmul(pts, cam_extr[:3, :3].transpose()) + cam_extr[:3, 3]
pts_proj = np.matmul(pts_proj, cam_intr.transpose())
pts_proj = pts_proj / pts_proj[:, 2:]
# pts_proj = np.int32(np.round(pts_proj[:, :2]))
# img = np.zeros([image_size[1], image_size[0], 3], dtype=np.uint8)
# for p in pts_proj[::50]:
# p = np.clip(p, 0, image_size[0] - 1)
# cv.circle(img, (int(p[0]), int(p[1])), 2, (0, 255, 0), -1)
# cv.imshow('img', img)
pts_min, pts_max = np.min(pts_proj, axis=0), np.max(pts_proj, axis=0)
pts_center = (pts_min + pts_max) // 2
pts_size = np.max(pts_max - pts_min)
tgt_pts_size = 350
tgt_image_size = 512
zoom_scale = tgt_pts_size / pts_size
cam_intr_zoom = np.copy(cam_intr)
cam_intr_zoom[:2] *= zoom_scale
cam_intr_zoom[0, 2] = cam_intr_zoom[0, 2] - (pts_center[0]*zoom_scale - tgt_image_size/2)
cam_intr_zoom[1, 2] = cam_intr_zoom[1, 2] - (pts_center[1]*zoom_scale - tgt_image_size/2)
head_cam_intr_zoom.append(cam_intr_zoom)
head_zoom_center.append(pts_center)
head_zoom_scale.append(zoom_scale)
# pts_proj = np.matmul(pts, cam_extr[:3, :3].transpose()) + cam_extr[:3, 3]
# pts_proj = np.matmul(pts_proj, cam_intr_zoom.transpose())
# pts_proj = pts_proj / pts_proj[:, 2:]
# pts_proj = np.int32(np.round(pts_proj[:, :2]))
# img = np.zeros([512, 512, 3], dtype=np.uint8)
# for p in pts_proj[::50]:
# p = np.clip(p, 0, image_size[0] - 1)
# cv.circle(img, (int(p[0]), int(p[1])), 2, (0, 255, 0), -1)
# cv.imshow('img_zoom', img)
# cv.waitKey()
np.savez(os.path.join(os.path.dirname(body_head_blending_param_path), 'head_zoomin_render_param.npz'),
cam_extr=head_cam_extr, cam_intr=head_cam_intr, image_size=image_size,
cam_intr_zoom=head_cam_intr_zoom, zoom_image_size=[tgt_image_size, tgt_image_size],
zoom_center=head_zoom_center,
zoom_scale=head_zoom_scale,
head_pose=head_pose, head_scale=head_scale, head_color_bw=head_color_bw)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
"""
body_gaussian_root_dir, ref_head_gaussian_path, ref_head_param_path, render_cam_fpath,
body_head_blending_param_path
"""
parser.add_argument('--body_gaussian_root_dir', type=str)
parser.add_argument('--ref_head_gaussian_path', type=str)
parser.add_argument('--ref_head_param_path', type=str)
parser.add_argument('--render_cam_fpath', type=str)
parser.add_argument('--body_head_blending_param_path', type=str)
args = parser.parse_args()
calc_offline_rendering_param(
args.body_gaussian_root_dir,
args.ref_head_gaussian_path,
args.ref_head_param_path,
args.render_cam_fpath,
args.body_head_blending_param_path
)
"""
python calc_offline_rendering_param.py ^
--body_gaussian_root_dir ./AnimatableGaussians/test_results/huawei0425/checkpoints/AMASS__test_poses_ours_front_view/batch_750000/pca_20_sigma_2.00/ ^
--ref_head_gaussian_path ./Gaussian-Head-Avatar/results/reenactment/huawei0425_self/posed_gaussians/000000.ply ^
--ref_head_param_path ./Gaussian-Head-Avatar/results/reenactment/huawei0425_self/params/000000_param.npz ^
--render_cam_fpath ./AnimatableGaussians/test_results/huawei0425/checkpoints/AMASS__test_poses_ours_front_view/batch_750000/pca_20_sigma_2.00/cameras.json ^
--body_head_blending_param_path ./data/body_face_stitching_sr/body_head_blending_param.npz
"""
|