File size: 4,257 Bytes
01fdb75
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import copy
import timm
from torch.nn import Parameter

from src.utils.no_grad import no_grad, freeze_model
from typing import Callable, Iterator, Tuple
from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
from torchvision.transforms import Normalize
from src.diffusion.base.training import *
from src.diffusion.base.scheduling import BaseScheduler
import lpips

def inverse_sigma(alpha, sigma):
    return 1/sigma**2
def snr(alpha, sigma):
    return alpha/sigma
def minsnr(alpha, sigma, threshold=5):
    return torch.clip(alpha/sigma, min=threshold)
def maxsnr(alpha, sigma, threshold=5):
    return torch.clip(alpha/sigma, max=threshold)
def constant(alpha, sigma):
    return 1


def time_shift_fn(t, timeshift=1.0):
    return t/(t+(1-t)*timeshift)


class REPATrainer(BaseTrainer):
    def __init__(
            self,
            scheduler: BaseScheduler,
            loss_weight_fn:Callable=constant,
            feat_loss_weight: float=0.5,
            lognorm_t=False,
            timeshift=1.0,
            encoder:nn.Module=None,
            align_layer=8,
            proj_denoiser_dim=256,
            proj_hidden_dim=256,
            proj_encoder_dim=256,
            P_mean=-0.8,
            P_std=0.8,
            t_eps=0.05,
            lpips_weight: float=1.0,
            *args,
            **kwargs
    ):
        super().__init__(*args, **kwargs)
        self.lognorm_t = lognorm_t
        self.scheduler = scheduler
        self.timeshift = timeshift
        self.loss_weight_fn = loss_weight_fn
        self.feat_loss_weight = feat_loss_weight
        self.align_layer = align_layer
        self.encoder = encoder
        freeze_model(self.encoder)
        
        self.lpips_loss_fn = lpips.LPIPS(net='vgg').eval()
        self.lpips_loss_fn.compile()
        freeze_model(self.lpips_loss_fn)
       
        self.proj = nn.Sequential(
            nn.Sequential(
                nn.Linear(proj_denoiser_dim, proj_hidden_dim),
                nn.SiLU(),
                nn.Linear(proj_hidden_dim, proj_hidden_dim),
                nn.SiLU(),
                nn.Linear(proj_hidden_dim, proj_encoder_dim),
            )
        )
        self.P_mean = P_mean
        self.P_std = P_std
        self.t_eps = t_eps
        self.lpips_weight = lpips_weight

    def _impl_trainstep(self, net, ema_net, solver, x, y, metadata=None):
        raw_images = metadata["raw_image"]
        batch_size, c, height, width = x.shape
        self.lpips_loss_fn.eval()
        if self.lognorm_t:
            base_t = (torch.randn(batch_size, device=x.device, dtype=torch.float32)*self.P_std+self.P_mean).sigmoid()
        else:
            base_t = torch.rand((batch_size), device=x.device, dtype=torch.float32)
        t = time_shift_fn(base_t, self.timeshift) #.to(x.dtype)
        noise = torch.randn_like(x)
        alpha = self.scheduler.alpha(t)
        dalpha = self.scheduler.dalpha(t)
        sigma = self.scheduler.sigma(t)
        dsigma = self.scheduler.dsigma(t)

        x_t = alpha * x + noise * sigma

        v_t = (x - x_t) / (1 - t.view(-1, 1, 1, 1)).clamp_min(self.t_eps)

        # v_t = dalpha * x + dsigma * noise

        pred_img, src_feature = net(x_t, t, y, return_layer=self.align_layer)
        src_feature = self.proj(src_feature)
        out = (pred_img - x_t) / (1 - t.view(-1, 1, 1, 1)).clamp_min(self.t_eps) # compute v from pred x

        with torch.no_grad():
            dst_feature = self.encoder(raw_images)
        cos_sim = torch.nn.functional.cosine_similarity(src_feature, dst_feature, dim=-1)
        cos_loss = 1 - cos_sim

        weight = self.loss_weight_fn(alpha, sigma)
        fm_loss = weight*(out - v_t)**2
        
        lpips_loss = self.lpips_loss_fn(pred_img, x)
       
        out = dict(
            fm_loss=fm_loss.mean(),
            cos_loss=cos_loss.mean(),
            lpips_loss=lpips_loss.mean(),
            loss=fm_loss.mean() + self.feat_loss_weight*cos_loss.mean() + self.lpips_weight*lpips_loss.mean(),
        )

        return out

    def state_dict(self, *args, destination=None, prefix="", keep_vars=False):
        self.proj.state_dict(
            destination=destination,
            prefix=prefix + "proj.",
            keep_vars=keep_vars)