-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_color_cnn.py
176 lines (136 loc) · 5.23 KB
/
train_color_cnn.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
#%%
import vapoursynth as vs
core = vs.core
core.std.LoadPlugin("/usr/lib/vapoursynth/libvsrawsource.so")
core.std.LoadPlugin("/usr/lib/vapoursynth/libakarin.so")
core.std.LoadPlugin("/usr/lib/vapoursynth/bestsource.so")
core.std.LoadPlugin("/usr/lib/vapoursynth/libresize2.so")
core.std.LoadPlugin("/usr/lib/vapoursynth/libcolorbars.so")
core.std.LoadPlugin("/usr/lib/vapoursynth/libvslsmashsource.so")
#%%
import json
import functools
from pathlib import Path
from vstools import remap_frames, vs, initialize_clip, set_output, depth, padder, Matrix, Transfer, Primaries, ColorRange,split,join
import random
import numpy as np
from matplotlib import pyplot as plt
from ldzeug2.vsnn import fill_train_buffer,cut_to_rndm_frames,cache_all_frames, get_rndm_frames,interleave_clips,load_random_train_frame_from_vnode, load_random_train_frame_from_vnode_idx_given
from ldzeug2.colorencoder import modulate_fields
import numpy as np
import matplotlib.pyplot as plt
import vskernels as vke
exec(open("clips.py","rt").read())
clips: list[vs.VideoNode] = out
og = interleave_clips(clips)
on_fields = True
kernel_in = vke.Bicubic()
#kernel_out = vke.Gaussian(sigma=0.5)
kernel_out = kernel_in
ccnt = 60
if on_fields:
og_in = kernel_in.scale(og,760,486,format=vs.YUV444P16).std.SeparateFields(tff=True)
og_out = kernel_out.scale(og,760,486,format=vs.YUV444P16).std.SeparateFields(tff=True)
frms = get_rndm_frames(og)
#from vstools import remap_frames
og_in = remap_frames(og_in,frms[:ccnt])
og_out = remap_frames(og_out,frms[:ccnt])
else:
assert False
og = kernel.scale(og,760,486,format=vs.YUV444P16)
remaped = cut_to_rndm_frames(og,ccnt).std.SeparateFields(tff=True)
remaped_in = core.std.Interleave([og_in,og_in])
remaped_out = core.std.Interleave([og_out,og_out])
modulated_in = modulate_fields(remaped_in)
modulated_out = modulate_fields(remaped_out)
train_input = join([modulated_in.tbc_out, modulated_in.i_carier,modulated_in.q_carier])
train_output = join([modulated_out.luma_out,modulated_out.i_hp,modulated_out.q_hp])
if on_fields:
pass
else:
train_output = train_output.std.DoubleWeave(tff=True)[::2]
train_input = train_input.std.DoubleWeave(tff=True)[::2]
print("caching")
cache_all_frames(train_input)
cache_all_frames(train_output)
#%%
import pyqtgraph as pg
%gui qt5
pg.image(np.array([
np.array(train_output.get_frame(0)[0]).transpose(),
np.array(train_input.get_frame(0)[0]).transpose(),
]))
#%%
import torch
import os
torch.set_default_device('cuda')
mdlpth = "/tmp/rr46.pth"
from ldzeug2.colorcnn_trch import FullModel2
model = FullModel2(num_feat=64,num_conv=16)
if os.path.exists(mdlpth):
model.load_state_dict(torch.load(mdlpth))
i = 0
init_lr = 1e-4
optimizer = torch.optim.Adam(model.parameters(), lr=init_lr,betas=[0.9, 0.99])
#optimizer = torch.optim.SGD(model.parameters(), lr=init_lr, momentum=0.0)
loss_fn = torch.nn.MSELoss()
#%%
gt_size = 64
batch_cnt = 12
save_it = 1000
tnsrss_in = torch.ones((batch_cnt,3,gt_size,gt_size))
tnsrss_out = torch.ones((batch_cnt,3,gt_size,gt_size))
while True:
frame_num, f_width,f_height,hq,lq = load_random_train_frame_from_vnode_idx_given((i % len(train_output) // 2) * 2+0,train_output,train_input)
lq = torch.Tensor(np.expand_dims(lq,0)).cuda()
hq = torch.Tensor(np.expand_dims(hq,0)).cuda()
fill_train_buffer(tnsrss_in,tnsrss_out,lq,hq,f_width,f_height,gt_size,batch_cnt//2,0)
frame_num, f_width,f_height,hq,lq = load_random_train_frame_from_vnode_idx_given((i % len(train_output) // 2) * 2+1,train_output,train_input)
lq = torch.Tensor(np.expand_dims(lq,0)).cuda()
hq = torch.Tensor(np.expand_dims(hq,0)).cuda()
fill_train_buffer(tnsrss_in,tnsrss_out,lq,hq,f_width,f_height,gt_size,batch_cnt//2,batch_cnt//2)
optimizer.zero_grad()
model_outputs = model(tnsrss_in)
loss = loss_fn(model_outputs, tnsrss_out)
loss.backward()
optimizer.step()
if (i % save_it) == 0:
torch.save(model.state_dict(),mdlpth)
losnow = loss.detach().cpu().numpy()
print(i,losnow)
if i % save_it == 0:
mdll = model_outputs.detach().cpu().numpy()[0,0]
hql = tnsrss_out.detach().cpu().numpy()[0,0]
plt.subplot(241)
plt.imshow(mdll)
plt.title(f"model luma")
#plt.show()
plt.subplot(242)
plt.imshow(hql)
plt.title("hq luma")
#plt.show()
plt.subplot(243)
plt.imshow(tnsrss_in.detach().cpu().numpy()[0,0])
plt.title("lq luma")
plt.subplot(244)
plt.imshow((torch.abs(model_outputs - tnsrss_out)).detach().cpu().numpy()[0,0])
plt.title("err luma")
mdll = model_outputs.detach().cpu().numpy()[0,1]
hql = tnsrss_out.detach().cpu().numpy()[0,1]
plt.subplot(245)
plt.imshow(mdll)
plt.title(f"model i")
#plt.show()
plt.subplot(246)
plt.imshow(hql)
plt.title("hq i")
#plt.show()
plt.subplot(247)
plt.imshow(tnsrss_in.detach().cpu().numpy()[0,1])
plt.title("i mlt")
plt.subplot(248)
plt.imshow((torch.abs(model_outputs - tnsrss_out)).detach().cpu().numpy()[0,1])
plt.title("err i")
plt.show()
i += 1
# %%