-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
494 lines (367 loc) · 15 KB
/
train.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
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
import sys
import torch
import torch.nn as nn
import torch.nn.functional as F
from tqdm import tqdm
import os
import numpy as np
import random
import torch.nn.init as init
import struct
from dataclasses import dataclass, fields
import math
from torch.utils.data import Dataset, DataLoader
import torchvision.transforms as transforms
SIZE_K = 3 #kernel_size in conv
SIZE_P = 1 #padding_size in conv
@dataclass
class Config:
width: int = 16
height: int = 16
n_in: int = 3 # in_channels, dfault to 3
n_feature: int = 64 # out_chanels
n_cfeature: int = 5 #contxt embeding size
cfg = Config(**dict(
n_feature = 64,
))
class GELU(nn.Module):
def forward(self, x):
return 0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 * torch.pow(x, 3))))
def serialize(t: torch.Tensor):
d = t.detach().cpu().view(-1).numpy().astype(np.float32)
b = struct.pack(f'{len(d)}f', *d)
return b
class ExportMixin:
def export (self, f):
for layer in self.layers_export:
if hasattr(layer, 'export'):
layer.export(f)
elif isinstance(layer, nn.Module):
if hasattr(layer, 'weight'):
f.write(serialize(layer.weight))
if hasattr(layer, 'bias'):
f.write(serialize(layer.bias))
class ResidualConvBlock(nn.Module):
def __init__(self, n_in, n_out, is_res=False):
super().__init__()
self.is_res = is_res
self.same_channels = n_in == n_out
size_k = SIZE_K
size_p = SIZE_P
conv1_core = nn.Conv2d(n_in, n_out, size_k, 1, size_p)
conv2_core = nn.Conv2d(n_out, n_out, size_k, 1, size_p)
conv3_core = nn.Conv2d(n_in, n_out, kernel_size=1, stride=1, padding=0) if is_res and (n_in != n_out) else None
self.conv1_core = conv1_core
self.conv2_core = conv2_core
self.conv3_core = conv3_core
self.conv1 = nn.Sequential(
conv1_core, # 3x3 kernel with stride 1 and padding 1
nn.BatchNorm2d(n_out), # Batch normalization
GELU(), # GELU activation function
# nn.GELU(), # GELU activation function
)
# Second convolutional layer
self.conv2 = nn.Sequential(
conv2_core, # 3x3 kernel with stride 1 and padding 1
nn.BatchNorm2d(n_out), # Batch normalization
GELU(), # GELU activation functions
)
def forward(self, x):
x1 = self.conv1(x)
x2 = self.conv2(x1)
if self.is_res:
if self.same_channels:
out = x + x2
else:
shortcut = self.conv3_core(x)
out = shortcut + x2
return out / 1.414
else:
return x2
# Method to get the number of output channels for this block
def get_out_channels(self):
return self.conv2[0].out_channels
# Method to set the number of output channels for this block
def set_out_channels(self, out_channels):
self.conv1[0].out_channels = out_channels
self.conv2[0].in_channels = out_channels
self.conv2[0].out_channels = out_channels
def export(self, f):
f.write(serialize(self.conv1_core.weight))
f.write(serialize(self.conv1_core.bias))
f.write(serialize(self.conv2_core.weight))
f.write(serialize(self.conv2_core.bias))
if self.conv3_core is not None:
f.write(serialize(self.conv3_core.weight))
f.write(serialize(self.conv3_core.bias))
class UnetDown(nn.Module, ExportMixin):
def __init__(self, n_in, n_out):
super().__init__()
# Create a list of layers for the downsampling block
# Each block consists of two ResidualConvBlock layers, followed by a MaxPool2d layer for downsampling
block1 = ResidualConvBlock(n_in, n_out)
block2 = ResidualConvBlock(n_out, n_out)
self.layers_export = [block1, block2]
# layers = [block1, block2]
layers = [block1, block2, nn.MaxPool2d(2)]
# Use the layers to create a sequential model
self.model = nn.Sequential(*layers)
def forward(self, x):
# Pass the input through the sequential model and return the output
return self.model(x)
class UnetUp(nn.Module, ExportMixin):
def __init__(self, n_in, n_out):
super().__init__()
block0 = nn.ConvTranspose2d(n_in, n_out, 2, 2)
block1 = ResidualConvBlock(n_out, n_out)
block2 = ResidualConvBlock(n_out, n_out)
self.layers_export = [block0, block1, block2]
layers = self.layers_export
# Use the layers to create a sequential model
self.model = nn.Sequential(*layers)
def forward(self, x, skip):
# Concatenate the input tensor x with the skip connection tensor along the channel dimension
x = torch.cat((x, skip), 1)
x = self.model(x)
return x
class EmbedFC(nn.Module, ExportMixin):
def __init__(self, n_in, n_out):
super().__init__()
self.n_in = n_in
self.n_out = n_out
block1 = nn.Linear(n_in, n_out)
block2 = nn.Linear(n_out, n_out)
self.layers_export = [block1, block2]
layers = [block1, GELU(), block2]
self.model = nn.Sequential(*layers)
def forward(self, x):
x = x.view(-1, self.n_in)
return self.model(x)
class Out(nn.Module, ExportMixin):
def __init__(self, n_in, n_out):
super().__init__()
size_k = SIZE_K
size_p = SIZE_P
conv1 = nn.Conv2d(2 * n_out, n_out, size_k, 1, size_p)
conv2 = nn.Conv2d(n_out, n_in, size_k, 1, size_p)
self.layers_export = [conv1, conv2]
layers = [conv1, nn.GroupNorm(8, n_out), nn.ReLU(), conv2]
self.model = nn.Sequential(*layers)
def forward(self, x):
return self.model(x)
class ContextUnet(nn.Module, ExportMixin):
def __init__(self, n_in, n_feat, n_cfeat=5, image_size=16):
super().__init__()
self.n_in = n_in
self.n_feat = n_feat
self.n_cfeat = n_cfeat
self.image_size = image_size
block_in = ResidualConvBlock(n_in, n_feat, True)
down1 = UnetDown(n_feat, n_feat)
down2 = UnetDown(n_feat, 2 * n_feat)
to_vec = nn.Sequential(nn.AvgPool2d((4)), GELU())
timeembed1 = EmbedFC(1, 2 * n_feat)
timeembed2 = EmbedFC(1, n_feat)
contextembed1 = EmbedFC(n_cfeat, 2 * n_feat)
contextembed2 = EmbedFC(n_cfeat, n_feat)
up0_size_k = image_size //4
up0_size_s = image_size //4
up0_core = nn.ConvTranspose2d(2 * n_feat, 2 * n_feat, up0_size_k, up0_size_s)
up0 = nn.Sequential(up0_core, nn.GroupNorm(8, 2 * n_feat), nn.ReLU())
up1 = UnetUp(4 * n_feat, n_feat)
up2 = UnetUp(2 * n_feat, n_feat)
block_out = Out(n_in, n_feat)
self.layers_export = [
block_in,
down1, down2,
timeembed1, timeembed2, contextembed1, contextembed2,
up0_core, up1, up2,
block_out
]
self.block_in = block_in
self.down1 = down1
self.down2 = down2
self.to_vec = to_vec
self.timeembed1 = timeembed1
self.timeembed2 = timeembed2
self.contextembed1 = contextembed1
self.contextembed2 = contextembed2
self.up0 = up0
self.up1 = up1
self.up2 = up2
self.block_out = block_out
def forward(self, x, t, c):
x = self.block_in(x)
down1 = self.down1(x)
down2 = self.down2(down1)
hiddenvec = self.to_vec(down2)
cemb1 = self.contextembed1(c).view(-1, cfg.n_feature * 2, 1, 1) # (batch, 2*n_feat, 1,1)
temb1 = self.timeembed1(t).view(-1, cfg.n_feature * 2, 1, 1)
cemb2 = self.contextembed2(c).view(-1, cfg.n_feature, 1, 1)
temb2 = self.timeembed2(t).view(-1, cfg.n_feature, 1, 1)
up0 = self.up0(hiddenvec)
up0 = cemb1 * up0 + temb1
up1 = self.up1(up0, down2)
up1 = cemb2 * up1 + temb2
up2 = self.up2(up1, down1)
out = self.block_out(torch.cat((up2, x), 1))
return out
class CustomDataset(Dataset):
def __init__(self, sfilename, lfilename, transform, null_context=False):
self.sprites = np.load(sfilename)
self.slabels = np.load(lfilename)
print(f"sprite shape: {self.sprites.shape}")
print(f"labels shape: {self.slabels.shape}")
self.transform = transform
self.null_context = null_context
self.sprites_shape = self.sprites.shape
self.slabel_shape = self.slabels.shape
# Return the number of images in the dataset
def __len__(self):
return len(self.sprites)
# Get the image and label at a given index
def __getitem__(self, idx):
# Return the image and label as a tuple
if self.transform:
image = self.transform(self.sprites[idx])
if self.null_context:
label = torch.tensor(0).to(torch.int64)
else:
label = torch.tensor(self.slabels[idx]).to(torch.int64)
return (image, label)
def getshapes(self):
# return shapes of data and labels
return self.sprites_shape, self.slabel_shape
transform = transforms.Compose([
transforms.ToTensor(), # from [0,255] to range [0.0,1.0]
transforms.Normalize((0.5,), (0.5,)) # range [-1,1]
])
timesteps = 500
beta1 = 1e-4
beta2 = 0.02
def train():
print("training...")
save_dir = os.path.abspath('./weights')
os.makedirs(save_dir, exist_ok=True)
device = torch.device("cuda:0" if torch.cuda.is_available() else torch.device('cpu'))
nn_model = ContextUnet(cfg.n_in, n_feat=cfg.n_feature, n_cfeat=cfg.n_cfeature, image_size=cfg.height).to(device)
# training hyperparameters
batch_size = 100
n_epoch = 32
lrate = 1e-3
b_t = (beta2 - beta1) * torch.linspace(0, 1, timesteps + 1, device=device) + beta1
# def2.2
a_t = 1 - b_t
# def2.3
ab_t = torch.cumsum(a_t.log(), dim=0).exp()
ab_t[0] = 1
# load dataset and construct optimizer
dataset = CustomDataset("./data/sprites_1788_16x16.npy", "./data/sprite_labels_nc_1788_16x16.npy", transform,
null_context=False)
dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=True, num_workers=1)
optim = torch.optim.Adam(nn_model.parameters(), lr=lrate)
def perturb_input(x, t, noise):
return ab_t.sqrt()[t, None, None, None] * x + (1 - ab_t[t, None, None, None]) * noise
nn_model.train()
for ep in range(n_epoch):
print(f'epoch {ep}')
# linearly decay learning rate
optim.param_groups[0]['lr'] = lrate*(1-ep/n_epoch)
pbar = tqdm(dataloader, mininterval=2 )
for x, c in pbar: # x: images
optim.zero_grad()
x = x.to(device)
c = c.to(torch.float32).to(device)
# perturb data
noise = torch.randn_like(x)
t = torch.randint(1, timesteps + 1, (x.shape[0],)).to(device)
# 公式2.5
x_pert = perturb_input(x, t, noise)
# use network to recover noise
pred_noise = nn_model(x_pert, t / timesteps, c)
# loss is mean squared error between the predicted and true noise
loss = F.mse_loss(pred_noise, noise)
loss.backward()
optim.step()
# save model periodically
if ep%4==0 or ep == int(n_epoch-1):
torch.save(nn_model.state_dict(), os.path.join(save_dir, f"ckpt_{ep}.pth"))
print(f"saved model ckpt_{ep}.pth")
# print("loading weights")
# nn_model.load_state_dict(torch.load(f"weights/context_model_31.pth", map_location=device))
# print("check point keys:")
# checkpoint = torch.load(f"weights/context_model_31.pth", map_location=device)
# ckeys = checkpoint.keys()
# # ckeys = [k for k in ckeys if k.startswith("init_conv")]
# print(ckeys)
# print(len(ckeys))
#
# print("==============================")
# print("model keys:")
# mkeys = nn_model.state_dict().keys()
# mkeys = [k for k in mkeys if k.startswith("block_in")]
#
# print(mkeys)
# print(len(mkeys))
def infer(timesteps=500, beta2=0.02, beta1=1e-4):
save_dir = os.path.abspath('./weights')
device = torch.device("cuda:0" if torch.cuda.is_available() else torch.device('cpu'))
nn_model = ContextUnet(cfg.n_in, n_feat=cfg.n_feature, n_cfeat=cfg.n_cfeature, image_size=cfg.height).to(device)
nn_model.load_state_dict(torch.load(f"{save_dir}/ckpt_31.pth", map_location=device))
nn_model.eval()
print("Loaded in Model")
b_t = (beta2 - beta1) * torch.linspace(0, 1, timesteps + 1, device=device) + beta1
# def2.2
a_t = 1 - b_t
# def2.3
ab_t = torch.cumsum(a_t.log(), dim=0).exp()
ab_t[0] = 1
def denoise_add_noise(x, t, pred_noise, z=None):
if z is None:
z = torch.randn_like(x)
noise = b_t.sqrt()[t] * z
# 公式2.13d
mean = (x - pred_noise * ((1 - a_t[t]) / (1 - ab_t[t]).sqrt())) / a_t[t].sqrt()
# 公式2.16
return mean + noise
@torch.no_grad()
def sample_ddpm(n_sample, save_rate=20):
# x_T ~ N(0, 1), sample initial noise
samples = torch.randn(n_sample, 3, cfg.height, cfg.width).to(device)
# array to keep track of generated steps for plotting
intermediate = []
for i in range(timesteps, 0, -1):
print(f'sampling timestep {i:3d}', end='\r')
# reshape time tensor
t = torch.tensor([i / timesteps])[:, None, None, None].to(device)
c = torch.tensor([0.0, 0.0, 0.0, 0.0, 1.0]).view(-1, cfg.n_cfeature).to(device)
# sample some random noise to inject back in. For i = 1, don't add back in noise
z = torch.randn_like(samples) if i > 1 else 0
## nn_model 即描述 epsilon_\theta 的神经网络
eps = nn_model(samples, t, c) # predict noise e_(x_t,t)
## 公式2.16
samples = denoise_add_noise(samples, i, eps, z)
if i % save_rate == 0 or i == timesteps or i < 8:
intermediate.append(samples.detach().cpu().numpy())
intermediate = np.stack(intermediate)
return samples, intermediate
samples, intermediate_ddpm = sample_ddpm(4)
return samples, intermediate_ddpm
def export():
print("exporting bin file for c...")
save_dir = os.path.abspath('./weights')
binpath = os.path.join(save_dir, 'ckpt.bin')
device = torch.device("cuda:0" if torch.cuda.is_available() else torch.device('cpu'))
nn_model = ContextUnet(cfg.n_in, n_feat=cfg.n_feature, n_cfeat=cfg.n_cfeature, image_size=cfg.height).to(device)
nn_model.load_state_dict(torch.load(f"{save_dir}/ckpt_31.pth", map_location=device))
nn_model.eval()
with open(binpath, 'wb') as f:
for field in fields(Config):
b = struct.pack('i', getattr(cfg, field.name))
f.write(b)
nn_model.export(f)
print(f"wrote to {binpath}")
pass
if __name__ == '__main__':
train()
export()