forked from zlai0/VideoAutoencoder
-
Notifications
You must be signed in to change notification settings - Fork 0
/
test_re10k.py
136 lines (119 loc) · 5.03 KB
/
test_re10k.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
import argparse
import os
import time
import logger
import numpy as np
import torch
from torch.utils.data import DataLoader
import torchvision.io as io
from parser import test_re10k_parser
import data.image_folder as D
import data.data_loader as DL
from models.autoencoder import *
from test_helper import *
from tqdm import tqdm
import warnings
warnings.filterwarnings("ignore", category=UserWarning, module="torch.nn.functional")
parser = test_re10k_parser()
args = parser.parse_args()
np.set_printoptions(precision=3)
def gettime():
# get GMT time in string
return time.strftime("%Y%m%d%H%M%S", time.gmtime())
def main():
if not os.path.isdir(args.savepath):
os.makedirs(args.savepath)
args.savepath = args.savepath+f'/test_re10k_{gettime()}'
log = logger.setup_logger(args.savepath + '/testing.log')
for key, value in sorted(vars(args).items()):
log.info(str(key) + ': ' + str(value))
TestData, _ = D.dataloader(args.dataset, 1, args.interval,
is_train=args.train_set, load_all_frames=True)
TestLoader = DataLoader(DL.ImageFloder(TestData, args.dataset),
batch_size=1, shuffle=False, num_workers=0)
# get auto-encoder
encoder_3d = Encoder3D(args)
encoder_traj = EncoderTraj(args)
rotate = Rotate(args)
decoder = Decoder(args)
# cuda
encoder_3d = nn.DataParallel(encoder_3d).cuda()
encoder_traj = nn.DataParallel(encoder_traj).cuda()
rotate = nn.DataParallel(rotate).cuda()
decoder = nn.DataParallel(decoder).cuda()
if args.resume:
if os.path.isfile(args.resume):
log.info("=> loading checkpoint '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
encoder_3d.load_state_dict(checkpoint['encoder_3d'])
encoder_traj.load_state_dict(checkpoint['encoder_traj'])
decoder.load_state_dict(checkpoint['decoder'])
rotate.load_state_dict(checkpoint['rotate'])
log.info("=> loaded checkpoint '{}'".format(args.resume))
else:
log.info("=> No checkpoint found at '{}'".format(args.resume))
log.info("=> Will start from scratch.")
else:
log.info('=> No checkpoint file. Start from scratch.')
start_full_time = time.time()
with torch.no_grad():
log.info('start testing.')
test(TestData, TestLoader, encoder_3d, encoder_traj, decoder, rotate, log)
log.info('full testing time = {:.2f} Minutes'.format((time.time() - start_full_time) / 60))
def test(data, dataloader, encoder_3d, encoder_traj, decoder, rotate, log):
_loss = AverageMeter()
video_limit = min(args.video_limit, len(dataloader))
frame_limit = args.frame_limit
for b_i, video_clips in tqdm(enumerate(dataloader)):
if b_i == video_limit: break
encoder_3d.eval()
encoder_traj.eval()
decoder.eval()
rotate.eval()
clip = video_clips[0,:frame_limit].cuda()
t, c, h, w = clip.size()
poses = get_poses(encoder_traj, clip)
trajectory = construct_trajectory(poses)
trajectory = trajectory.reshape(-1,12)
preds = []
for i in range(t-1):
if i == 0:
preds.append(clip[0:1])
scene_rep = encoder_3d(video_clips[:, 0])
scene_index = 0
elif i % args.reinit_k == 0:
# reinitialize 3d voxel
scene_rep = encoder_3d(pred)
scene_index = i
clip_in = torch.stack([clip[scene_index], clip[i+1]])
pose = get_pose_window(encoder_traj, clip_in)
z = euler2mat(pose[1:])
rot_codes = rotate(scene_rep, z)
output = decoder(rot_codes)
pred = F.interpolate(output, (h, w), mode='bilinear')
pred = torch.clamp(pred, 0, 1)
preds.append(pred)
# output
synth_save_dir = os.path.join(args.savepath, f"Videos")
os.makedirs(synth_save_dir, exist_ok=True)
preds = torch.cat(preds,dim=0)
pred = (preds.permute(0,2,3,1) * 255).byte().cpu()
io.write_video(synth_save_dir+f'/video_{b_i}_pred.mp4', pred, 6)
vid = (clip.permute(0,2,3,1) * 255).byte().cpu()
io.write_video(synth_save_dir+f'/video_{b_i}_true.mp4', vid, 6)
pose_save_dir = os.path.join(args.savepath, f"Poses")
os.makedirs(pose_save_dir, exist_ok=True)
true_camera_file = os.path.dirname(data[b_i][0]).replace('dataset_square', 'RealEstate10K')+'.txt'
with open(true_camera_file) as f:
f.readline() # remove line 0
poses = np.loadtxt(f)
camera = poses[:,7:].reshape([-1,12])[:len(trajectory)]
with open(pose_save_dir+f'/video_{b_i}_pred.txt','w') as f:
lines = [' '.join(map(str,y))+'\n' for y in trajectory.tolist()]
f.writelines(lines)
with open(pose_save_dir+f'/video_{b_i}_true.txt','w') as f:
lines = [' '.join(map(str,y))+'\n' for y in camera]
f.writelines(lines)
print()
if __name__ == '__main__':
main()