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
|
## -*- coding: utf-8 -*-
import tensorflow as tf
import numpy as np
import time
import glob
import scipy
import argparse
import os
import subprocess
from random import choice
from PIL import Image
from utils import LoadImage, DownSample, AVG_PSNR, depth_to_space_3D, DynFilter3D, LoadParams
from nets import FR_16L, FR_28L, FR_52L
parser = argparse.ArgumentParser()
parser.add_argument('--L', metavar='L', type=int, default=28, help='Network depth: One of 16, 28, 52')
parser.add_argument('--T', metavar='T', default='L', help='Input type: L(Low-resolution) or G(Ground-truth)')
parser.add_argument('--dataset', default=None, help='Name of dataset')
parser.add_argument('--in_dir', metavar='in_dir', default=None, help='Directory to process')
parser.add_argument('--out_dir', metavar='out_dir', default='./results/test', help='Directory to output to')
parser.add_argument('--network_dir', default='.', help='Path to networks')
parser.add_argument('--mov_from_dirs', action='store_true')
args = parser.parse_args()
# Size of input temporal radius
T_in = 7
# Upscaling factor
R = 4
# Selecting filters and residual generating network
if args.L == 16:
FR = FR_16L
elif args.L == 28:
FR = FR_28L
elif args.L == 52:
FR = FR_52L
else:
print('Invalid network depth: {} (Must be one of 16, 28, 52)'.format(args.L))
exit(1)
if not(args.T == 'L' or args.T =='G'):
print('Invalid input type: {} (Must be L(Low-resolution) or G(Ground-truth))'.format(args.T))
exit(1)
def process_dir(dir):
dir_partz = dir.split('/')
dataset = args.dataset = dir_partz[-2]
part = dir_partz[-1]
tag = '_'.join([dataset, str(args.L) + 'L', part])
out_path = os.path.join(args.out_dir, part)
render_path = os.path.join(args.out_dir, 'renders')
print('process dir {}, out path: {}'.format(dir, out_path))
os.makedirs(out_path, exist_ok=True)
if args.mov_from_dirs:
os.makedirs(render_path, exist_ok=True)
dir_frames = sorted(glob.glob(os.path.join(dir, '*.png')))
if not len(dir_frames):
print("{}: no frames!".format(dir))
return
# print(dir_frames)
frames = []
for f in dir_frames:
frames.append(LoadImage(f))
frames = np.asarray(frames)
frames_padded = np.lib.pad(frames, pad_width=((T_in//2,T_in//2),(0,0),(0,0),(0,0)), mode='constant')
for i in range(frames.shape[0]):
if (i % 100) == 1:
print('Scene {}: Frame {}/{} processing'.format(tag, i+1, frames.shape[0]))
in_L = frames_padded[i:i+T_in] # select T_in frames
in_L = in_L[np.newaxis,:,:,:,:]
out_H = sess.run(GH, feed_dict={L: in_L, is_train: False})
out_H = np.clip(out_H, 0, 1)
Image.fromarray(np.around(out_H[0,0]*255).astype(np.uint8)).save('{}/frame_{:05d}.png'.format(out_path, i+1))
if args.mov_from_dirs:
subprocess.call([
'ffmpeg',
'-hide_banner',
'-i', os.path.join(out_path, 'frame_%05d.png'),
'-y',
'-c:v', 'libx264',
'-preset', 'slow',
'-crf', '19',
'-vf', 'fps=25',
'-pix_fmt', 'yuv420p',
os.path.join(render_path, tag + '.mp4')
])
def G(x, is_train):
# shape of x: [B,T_in,H,W,C]
# Generate filters and residual
# Fx: [B,1,H,W,1*5*5,R*R]
# Rx: [B,1,H,W,3*R*R]
Fx, Rx = FR(x, is_train)
x_c = []
for c in range(3):
t = DynFilter3D(x[:,T_in//2:T_in//2+1,:,:,c], Fx[:,0,:,:,:,:], [1,5,5]) # [B,H,W,R*R]
t = tf.depth_to_space(t, R) # [B,H*R,W*R,1]
x_c += [t]
x = tf.concat(x_c, axis=3) # [B,H*R,W*R,3]
x = tf.expand_dims(x, axis=1)
Rx = depth_to_space_3D(Rx, R) # [B,1,H*R,W*R,3]
x += Rx
return x
# Gaussian kernel for downsampling
def gkern(kernlen=13, nsig=1.6):
import scipy.ndimage.filters as fi
# create nxn zeros
inp = np.zeros((kernlen, kernlen))
# set element at the middle to one, a dirac delta
inp[kernlen//2, kernlen//2] = 1
# gaussian-smooth the dirac, resulting in a gaussian filter mask
return fi.gaussian_filter(inp, nsig)
h = gkern(13, 1.6) # 13 and 1.6 for x4
h = h[:,:,np.newaxis,np.newaxis].astype(np.float32)
# Network
H = tf.placeholder(tf.float32, shape=[None, T_in, None, None, 3])
L_ = DownSample(H, h, R)
L = L_[:,:,2:-2,2:-2,:] # To minimize boundary artifact
is_train = tf.placeholder(tf.bool, shape=[]) # Phase ,scalar
with tf.variable_scope('G') as scope:
GH = G(L, is_train)
params_G = [v for v in tf.global_variables() if v.name.startswith('G/')]
# Session
config = tf.ConfigProto()
config.gpu_options.allow_growth=True
with tf.Session(config=config) as sess:
tf.global_variables_initializer().run()
# Load parameters
LoadParams(sess, [params_G], in_file=os.path.join(args.network_dir, 'params_{}L_x{}.h5'.format(args.L, R)))
if args.T == 'L':
# Test using Low-resolution videos
if args.in_dir:
for dir in sorted(glob.glob(os.path.join(args.in_dir, '*'))):
process_dir(dir)
else:
for dir in sorted(glob.glob('./inputs/L/*')):
process_dir(dir)
elif args.T == 'G':
# Test using GT videos
avg_psnrs = []
dir_inputs = glob.glob('./inputs/G/*')
for v in dir_inputs:
scene_name = v.split('/')[-1]
os.makedirs('./results/{}L/G/{}/'.format(args.L, scene_name), exist_ok=True)
dir_frames = glob.glob(v + '/*.png')
dir_frames.sort()
frames = []
for f in dir_frames:
frames.append(LoadImage(f))
frames = np.asarray(frames)
frames_padded = np.lib.pad(frames, pad_width=((T_in//2,T_in//2),(0,0),(0,0),(0,0)), mode='constant')
frames_padded = np.lib.pad(frames_padded, pad_width=((0,0),(8,8),(8,8),(0,0)), mode='reflect')
out_Hs = []
for i in range(frames.shape[0]):
print('Scene {}: Frame {}/{} processing'.format(scene_name, i+1, frames.shape[0]))
in_H = frames_padded[i:i+T_in] # select T_in frames
in_H = in_H[np.newaxis,:,:,:,:]
out_H = sess.run(GH, feed_dict={H: in_H, is_train: False})
out_H = np.clip(out_H, 0, 1)
Image.fromarray(np.around(out_H[0,0]*255).astype(np.uint8)).save('./results/{}L/G/{}/frame_{:05d}.png'.format(args.L, scene_name, i+1))
out_Hs.append(out_H[0, 0])
out_Hs = np.asarray(out_Hs)
avg_psnr = AVG_PSNR(((frames)*255).astype(np.uint8)/255.0, ((out_Hs)*255).astype(np.uint8)/255.0, vmin=0, vmax=1, t_border=2, sp_border=8)
avg_psnrs.append(avg_psnr)
print('Scene {}: PSNR {}'.format(scene_name, avg_psnr))
|
