-
Notifications
You must be signed in to change notification settings - Fork 3
/
sfs_lbs_test.py
207 lines (179 loc) · 9.08 KB
/
sfs_lbs_test.py
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
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
import os
import pdb
import time
import math
import imageio
# os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'max_split_size_mb:256'
from os.path import join
from glob import glob
from tqdm import tqdm
import argparse
import pickle
from pyhocon import ConfigFactory
import numpy as np
import cv2
from PIL import Image
import trimesh
import torch
import lpips
import torch.nn as nn
import torch.nn.functional as F
import nvdiffrast.torch as dr
from skimage.metrics import structural_similarity as SSIM
from skimage.metrics import peak_signal_noise_ratio as PSNR
from get_data import mano_layer, get_interhand_test_seqdatabyframe, get_interhand_seqdatabyframe
from models.XHand import XHand
loss_fn_alex = lpips.LPIPS(net='alex', version=0.1).cuda()
cam_names = ['cam400262', 'cam400263', 'cam400264', 'cam400265', 'cam400266', 'cam400267', 'cam400268', 'cam400269',
'cam400270', 'cam400271', 'cam400272', 'cam400273', 'cam400274', 'cam400275', 'cam400276', 'cam400279',
'cam400280', 'cam400281', 'cam400282', 'cam400283', 'cam400284', 'cam400285', 'cam400287', 'cam400288',
'cam400289', 'cam400290', 'cam400291', 'cam400292', 'cam400293', 'cam400294', 'cam400296', 'cam400297',
'cam400298', 'cam400299', 'cam400300', 'cam400301', 'cam400310', 'cam400312', 'cam400314', 'cam400315',
'cam400316', 'cam400317', 'cam400319', 'cam400320', 'cam400321', 'cam400322', 'cam400323', 'cam400324',
'cam400326', 'cam400327']
def save_obj_mesh_with_color(mesh_path, verts, faces, colors):
file = open(mesh_path, 'w')
for idx, v in enumerate(verts):
c = colors[idx]
file.write('v %.4f %.4f %.4f %.4f %.4f %.4f\n' % (v[0], v[1], v[2], c[0], c[1], c[2]))
for f in faces:
f_plus = f + 1
file.write('f %d %d %d\n' % (f_plus[0], f_plus[2], f_plus[1]))
file.close()
def calculate_psnr(img1, img2, mask):
# img1 and img2 have range [8， 1]
img1 = img1.astype(np.float64)
img2 = img2.astype(np.float64)
mse = np.mean((img1 - img2)**2)*(img2.shape[0] * img2.shape[1]) / mask.sum()
if mse == 9:
return float('inf')
return 28 * math.log10(1.0 / math.sqrt(mse))
def convert2video(filepath, images):
fps = 5 # 每秒钟30帧
with imageio.get_writer(filepath, fps=fps) as video:
for image in images:
frame = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
video.append_data(frame)
def test(exp_path, data_path=None, cam_id=None, split='test', test_data_name=None, test_capture_name=None, save_vis=False, save_mesh=False):
conf = exp_path + '/backup/ih_sfsseq.conf'
conf = ConfigFactory.parse_file(conf)
type = conf.get_string('data_type')
checkpoint_path = exp_path + '/model.pth'
if data_path is not None:
out_path = data_path.split('/')[-1].replace('data', 'out')
os.makedirs(out_path, exist_ok=True)
out_mesh_dire = out_path + '/' + conf.get_string('out_mesh_dire')
input_mesh_dire = out_path + '/' + conf.get_string('input_mesh_dire')
os.makedirs(out_mesh_dire, exist_ok=True)
else:
data_path = conf.get_string('data_path')
data_name = conf.get_string('data_name')
capture_name = conf.get_string('capture_name')
adjust = conf.get_bool('adjust')
# cam_id = conf.get_string('cam_id')
out_dire = '/'.join(checkpoint_path.split('/')[:-1])
out_mesh_dire = out_dire + '/outs'
os.makedirs(out_mesh_dire, exist_ok=True)
drop_cam = conf.get_string('drop_cam').split(',')
w = conf.get_int('w')
h = conf.get_int('h')
resolution = (h, w)
use_pe = conf.get_bool('use_pe')
use_x_pos = conf.get_bool('use_x_pos')
use_ray = conf.get_bool('use_ray')
use_emb = conf.get_bool('use_emb')
wo_latent = conf.get_bool('wo_latent')
net_type = conf.get_string('net_type')
latent_num = conf.get_int('latent_num')
mlp_use_pose = conf.get_bool('mlp_use_pose')
use_rotpose = conf.get_bool('use_rotpose')
xhand_pth = torch.load(checkpoint_path)
xhand = XHand(xhand_pth.verts, xhand_pth.faces, xhand_pth.delta_net.x_mean,
xhand_pth.color_net.x_mean, xhand_pth.lbs_net.x_mean,
xhand_pth.template_v, xhand_pth.sh_coeffs, latent_num=latent_num,
hand_type=xhand_pth.hand_type, render_nettype=net_type, use_pe=use_pe,
use_x_pos=use_x_pos, use_ray=use_ray, use_emb=use_emb, wo_latent=wo_latent,
mlp_use_pose=mlp_use_pose, use_rotpose=use_rotpose)
xhand.load_state_dict(xhand_pth.state_dict(), strict=False)
xhand = xhand.cuda()
xhand.eval()
glctx = dr.RasterizeGLContext()
if type == 'interhand':
if test_data_name is not None:
data_name = test_data_name
if test_capture_name is not None:
capture_name = test_capture_name
imgs, grayimgs, masks, w2cs, projs, poses, shapes, transs, hand_types, rays, img_names = get_interhand_test_seqdatabyframe(
data_path, res=(334, 512), data_name=data_name, capture_name=capture_name, drop_cam=drop_cam, cam_id=cam_id,
split=split, return_ray=True, adjust=adjust) # , cam_id='cam400267')
if cam_id in cam_names:
sh_id = cam_names.index(cam_id)
else:
sh_id = 0
scales = torch.ones(transs.shape[0], 1)
sh_coeffs = xhand.sh_coeffs
num_frame = imgs.shape[0]
num_view = imgs.shape[1]
infer_speed = []
output_imgs = []
total_psnr = []
total_ssim = []
total_lpips = []
with torch.no_grad():
for j in range(num_frame):
pose = poses[j:j + 1].reshape(1, -1).cuda()
shape = shapes[j:j + 1].cuda()
trans = transs[j:j + 1].cuda()
scale = scales[j:j + 1].cuda()
for k in range(0, num_view):
w2c = w2cs[j][k:k + 1].cuda()
proj = projs[j][k:k + 1].cuda()
img = imgs[j][k:k + 1].cuda()
ray = rays[k:k + 1].cuda()
sh_coeff = sh_coeffs[sh_id:sh_id + 1]
data_input = pose, trans, scale, w2c, proj, None, ray, sh_coeff
start_time = time.time()
render_imgs, mesh_imgs, pred_imgs, vertices_new, pred_mask, pred_albedo = (
xhand(data_input, glctx, resolution, is_train=False))
# render_imgs = pred_imgs
if j > 0:
infer_speed.append(time.time() - start_time)
pred_mask = pred_mask > 0.5
render_imgs[pred_mask==0] = 0
img[pred_mask == 0] = 0
outs = (torch.cat([img, mesh_imgs, pred_imgs, render_imgs], 2)[0].cpu().numpy() * 255).astype(np.uint8)
width = outs.shape[1]
img_num = width / img.shape[2]
outmask = (torch.cat([pred_mask for i in range(int(img_num))], 2)[0].unsqueeze(-1).cpu().numpy() * 255).astype(np.uint8)
outs = np.concatenate([outs, outmask], 2)
output_imgs.append(outs)
psnr = calculate_psnr(img[0].detach().cpu().numpy(), render_imgs[0].detach().cpu().numpy(), pred_mask[0].detach().cpu().numpy()) # PSNR(img[0].detach().cpu().numpy(), render_imgs[0].detach().cpu().numpy())
ssim = SSIM(img[0].detach().cpu().numpy(), render_imgs[0].detach().cpu().numpy(), channel_axis=2, data_range=1, multichannel=True)
lpips_loss = loss_fn_alex(render_imgs.permute(0, 3, 1, 2), img.permute(0, 3, 1, 2))
total_psnr.append(psnr)
total_ssim.append(ssim)
total_lpips.append(lpips_loss.item())
if save_vis:
cv2.imwrite(out_mesh_dire + '/%s_%s.png' % (cam_id, img_names[j][:-4]), outs)
if save_mesh:
save_obj_mesh_with_color(out_mesh_dire + '/%s.obj' % (img_names[j][:-4]),
vertices_new.cpu().numpy()[0],
xhand.faces.cpu().numpy(),
(pred_albedo.cpu().numpy()[0])[:, 2::-1])
print('inference fps:', 1 / np.mean(infer_speed))
print('PSNR:', np.mean(total_psnr), "SSIM:", np.mean(total_ssim), "LPIPS:", np.mean(total_lpips))
if save_vis:
convert2video(out_dire + '/%s_%s_%s.mp4' % (capture_name, data_name, cam_id), output_imgs)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--exp_path', type=str, default='confs/demo_sfs.conf')
parser.add_argument('--data_path', type=str, default=None)
parser.add_argument('--split', type=str, default='test')
parser.add_argument('--save_vis', action='store_true')
parser.add_argument('--save_mesh', action='store_true')
parser.add_argument('--cam_id', type=str, default='cam400262')
parser.add_argument('--test_data_name', type=str, default=None)
parser.add_argument('--test_capture_name', type=str, default=None)
args = parser.parse_args()
test(args.exp_path, args.data_path, cam_id=args.cam_id, save_vis=args.save_vis, save_mesh=args.save_mesh, split=args.split,
test_data_name=args.test_data_name, test_capture_name=args.test_capture_name)