File size: 7,946 Bytes
6f02465
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import pytorch_lightning as pl
import torch.nn.functional as F

import sys
sys.path.append('.')

from stable_diffusion.ldm.modules.diffusionmodules.model import Encoder, Decoder
from stable_diffusion.ldm.modules.distributions.distributions import DiagonalGaussianDistribution
from stable_diffusion.ldm.util import instantiate_from_config


class AutoencoderKL(pl.LightningModule):
    def __init__(self,
                 ddconfig,
                 lossconfig,  # torch.nn.Identity
                 embed_dim,  # embed_dim = 4
                 ckpt_path=None,
                 ignore_keys=[],
                 image_key="image",
                 colorize_nlabels=None,  # This is None
                 monitor=None,  # val/rec_loss
                 ):
        super().__init__()
        self.image_key = image_key  # 'image'

        # The encoder and decoder are reverse in the VQVAE
        # The encoder encodes the image to a latent space, and then transfer it to a Gaussian Distribution
        self.encoder = Encoder(**ddconfig)
        # Note, the output of the encoder is NOT directly fed into the decoder. The output channel size of the encoder is 2 * z_channel, as identified by the ddconfig['double_z']. This is becuase the output of the encoder is used to construct a Gaussian Distribution
        # The decoder decodes the latent space to an image
        self.decoder = Decoder(**ddconfig)

        # torch.nn.Identity
        self.loss = instantiate_from_config(lossconfig)  # Identity function

        # double_z = True.
        assert ddconfig["double_z"]

        # z_channels = 4
        self.quant_conv = torch.nn.Conv2d(2 * ddconfig["z_channels"], 2 * embed_dim, 1)
        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)

        # embed_dim = 4
        self.embed_dim = embed_dim

        # colorize_nlabels is None
        if colorize_nlabels is not None:
            assert type(colorize_nlabels)==int
            self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1))

        # monitor = val/rec_loss
        if monitor is not None:
            self.monitor = monitor

        # ckpt_path = None, the checkpoint loading of stable diffusion is conducted outside
        if ckpt_path is not None:
            self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys)

    def init_from_ckpt(self, path, ignore_keys=list()):
        sd = torch.load(path, map_location="cpu")["state_dict"]
        keys = list(sd.keys())
        for k in keys:
            for ik in ignore_keys:
                if k.startswith(ik):
                    print("Deleting key {} from state_dict.".format(k))
                    del sd[k]
        self.load_state_dict(sd, strict=False)
        print(f"Restored from {path}")

    def encode(self, x):
        # x: [bs, 3, 256, 256], h: [bs, 8, 32, 32]
        h = self.encoder(x)
        # serves as the mean and variance of the Gaussian distribution (halve the last dim)
        # moments: [bs, 8, 32, 32]
        moments = self.quant_conv(h)
        posterior = DiagonalGaussianDistribution(moments)
        return posterior

    def decode(self, z):
        # z: [bs, 4, 32, 32]
        z = self.post_quant_conv(z)
        # z: [bs, 4, 32, 32]
        dec = self.decoder(z)
        # dec: [bs, 3, 256, 256]
        return dec

    def forward(self, input, sample_posterior=True):
        posterior = self.encode(input)
        if sample_posterior:
            z = posterior.sample()  # a normal sampling
        else:
            z = posterior.mode()  # returns the mean
        dec = self.decode(z)
        return dec, posterior

    def get_input(self, batch, k):
        x = batch[k]
        if len(x.shape) == 3:
            x = x[..., None]
        x = x.permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format).float()
        return x

    def training_step(self, batch, batch_idx, optimizer_idx):  # in Stable Diffusion we use pretrained VAE and freeze it.
        inputs = self.get_input(batch, self.image_key)
        reconstructions, posterior = self(inputs)

        if optimizer_idx == 0:
            # train encoder+decoder+logvar
            aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
                                            last_layer=self.get_last_layer(), split="train")
            self.log("aeloss", aeloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
            self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=False)
            return aeloss

        if optimizer_idx == 1:
            # train the discriminator
            discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
                                                last_layer=self.get_last_layer(), split="train")

            self.log("discloss", discloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
            self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=False)
            return discloss

    def validation_step(self, batch, batch_idx):
        inputs = self.get_input(batch, self.image_key)
        reconstructions, posterior = self(inputs)
        aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, 0, self.global_step,
                                        last_layer=self.get_last_layer(), split="val")

        discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, 1, self.global_step,
                                            last_layer=self.get_last_layer(), split="val")

        self.log("val/rec_loss", log_dict_ae["val/rec_loss"])
        self.log_dict(log_dict_ae)
        self.log_dict(log_dict_disc)
        return self.log_dict

    def configure_optimizers(self):
        lr = self.learning_rate
        opt_ae = torch.optim.Adam(list(self.encoder.parameters())+
                                  list(self.decoder.parameters())+
                                  list(self.quant_conv.parameters())+
                                  list(self.post_quant_conv.parameters()),
                                  lr=lr, betas=(0.5, 0.9))
        opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
                                    lr=lr, betas=(0.5, 0.9))
        return [opt_ae, opt_disc], []

    def get_last_layer(self):
        return self.decoder.conv_out.weight

    @torch.no_grad()
    def log_images(self, batch, only_inputs=False, **kwargs):
        log = dict()
        x = self.get_input(batch, self.image_key)
        x = x.to(self.device)
        if not only_inputs:
            xrec, posterior = self(x)
            if x.shape[1] > 3:
                # colorize with random projection
                assert xrec.shape[1] > 3
                x = self.to_rgb(x)
                xrec = self.to_rgb(xrec)
            log["samples"] = self.decode(torch.randn_like(posterior.sample()))
            log["reconstructions"] = xrec
        log["inputs"] = x
        return log

    def to_rgb(self, x):
        assert self.image_key == "segmentation"
        if not hasattr(self, "colorize"):
            self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x))
        x = F.conv2d(x, weight=self.colorize)
        x = 2.*(x-x.min())/(x.max()-x.min()) - 1.
        return x


class IdentityFirstStage(torch.nn.Module):
    def __init__(self, *args, vq_interface=False, **kwargs):
        self.vq_interface = vq_interface  # TODO: Should be true by default but check to not break older stuff
        super().__init__()

    def encode(self, x, *args, **kwargs):
        return x

    def decode(self, x, *args, **kwargs):
        return x

    def quantize(self, x, *args, **kwargs):
        if self.vq_interface:
            return x, None, [None, None, None]
        return x

    def forward(self, x, *args, **kwargs):
        return x