File size: 7,393 Bytes
b5ce381
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from pytorch_lightning.callbacks import Callback
from pytorch_lightning.loggers import WandbLogger
import numpy as np
from pytorch_lightning.utilities import rank_zero_only
from typing import Union
import pytorch_lightning as pl
import os
from matplotlib import pyplot as plt
from sgm.util import exists, isheatmap
import torchvision
from PIL import Image
import torch
import wandb
from einops import rearrange


class ImageLogger(Callback):
    def __init__(
        self,
        batch_frequency,
        max_images,
        clamp=True,
        increase_log_steps=True,
        rescale=True,
        disabled=False,
        log_on_batch_idx=False,
        log_first_step=False,
        log_images_kwargs=None,
        log_before_first_step=False,
        enable_autocast=True,
    ):
        super().__init__()
        self.enable_autocast = enable_autocast
        self.rescale = rescale
        self.batch_freq = batch_frequency
        self.max_images = max_images
        self.log_steps = [2**n for n in range(int(np.log2(self.batch_freq)) + 1)]
        if not increase_log_steps:
            self.log_steps = [self.batch_freq]
        self.clamp = clamp
        self.disabled = disabled
        self.log_on_batch_idx = log_on_batch_idx
        self.log_images_kwargs = log_images_kwargs if log_images_kwargs else {}
        self.log_first_step = log_first_step
        self.log_before_first_step = log_before_first_step

    @rank_zero_only
    def log_local(
        self,
        save_dir,
        split,
        images,
        global_step,
        current_epoch,
        batch_idx,
        pl_module: Union[None, pl.LightningModule] = None,
    ):
        root = os.path.join(save_dir, "images", split)
        for k in images:
            if isheatmap(images[k]):
                fig, ax = plt.subplots()
                ax = ax.matshow(images[k].cpu().numpy(), cmap="hot", interpolation="lanczos")
                plt.colorbar(ax)
                plt.axis("off")

                filename = "{}_gs-{:06}_e-{:06}_b-{:06}.png".format(k, global_step, current_epoch, batch_idx)
                os.makedirs(root, exist_ok=True)
                path = os.path.join(root, filename)
                plt.savefig(path)
                plt.close()
                # TODO: support wandb
            else:
                grid = torchvision.utils.make_grid(images[k].squeeze(2), nrow=4)
                if self.rescale:
                    grid = (grid + 1.0) / 2.0  # -1,1 -> 0,1; c,h,w
                # print(grid.shape, grid.dtype, grid.min(), grid.max(), k)
                grid = rearrange(grid.squeeze(1), "c h w -> h w c")
                # grid = grid.transpose(0, 1).transpose(1, 2).squeeze(-1)
                grid = grid.numpy()
                grid = (grid * 255).astype(np.uint8)
                filename = "{}_gs-{:06}_e-{:06}_b-{:06}.png".format(k, global_step, current_epoch, batch_idx)
                path = os.path.join(root, filename)
                os.makedirs(os.path.split(path)[0], exist_ok=True)
                img = Image.fromarray(grid)
                img.save(path)
                if exists(pl_module):
                    assert isinstance(
                        pl_module.logger, WandbLogger
                    ), "logger_log_image only supports WandbLogger currently"
                    pl_module.logger.log_image(
                        key=f"{split}/{k}",
                        images=[
                            img,
                        ],
                        step=pl_module.global_step,
                    )

    @rank_zero_only
    def log_img(self, pl_module, batch, batch_idx, split="train"):
        check_idx = batch_idx if self.log_on_batch_idx else pl_module.global_step
        if (
            self.check_frequency(check_idx)
            and hasattr(pl_module, "log_images")  # batch_idx % self.batch_freq == 0
            and callable(pl_module.log_images)
            and
            # batch_idx > 5 and
            self.max_images > 0
        ):
            logger = type(pl_module.logger)
            is_train = pl_module.training
            if is_train:
                pl_module.eval()

            gpu_autocast_kwargs = {
                "enabled": self.enable_autocast,  # torch.is_autocast_enabled(),
                "dtype": torch.get_autocast_gpu_dtype(),
                "cache_enabled": torch.is_autocast_cache_enabled(),
            }
            with torch.no_grad(), torch.cuda.amp.autocast(**gpu_autocast_kwargs):
                images = pl_module.log_images(batch, split=split, **self.log_images_kwargs)

            for k in images:
                N = min(images[k].shape[0], self.max_images)
                if not isheatmap(images[k]):
                    images[k] = images[k][:N]
                if isinstance(images[k], torch.Tensor):
                    images[k] = images[k].detach().float().cpu()
                    if self.clamp and not isheatmap(images[k]):
                        images[k] = torch.clamp(images[k], -1.0, 1.0)

            self.log_local(
                pl_module.logger.save_dir,
                split,
                images,
                pl_module.global_step,
                pl_module.current_epoch,
                batch_idx,
                pl_module=pl_module if isinstance(pl_module.logger, WandbLogger) else None,
            )

            if is_train:
                pl_module.train()

    def check_frequency(self, check_idx):
        if check_idx:
            check_idx -= 1
        if ((check_idx % self.batch_freq) == 0 or (check_idx in self.log_steps)) and (
            check_idx > 0 or self.log_first_step
        ):
            try:
                self.log_steps.pop(0)
            except IndexError as e:
                print(e)
                pass
            return True
        return False

    @rank_zero_only
    def on_train_batch_end(self, trainer, pl_module, outputs, batch, batch_idx):
        if not self.disabled and (pl_module.global_step > 0 or self.log_first_step):
            self.log_img(pl_module, batch, batch_idx, split="train")

    @rank_zero_only
    def on_train_batch_start(self, trainer, pl_module, batch, batch_idx):
        if self.log_before_first_step and pl_module.global_step == 0:
            print(f"{self.__class__.__name__}: logging before training")
            self.log_img(pl_module, batch, batch_idx, split="train")

    @rank_zero_only
    def on_validation_batch_end(self, trainer, pl_module, outputs, batch, batch_idx, *args, **kwargs):
        if not self.disabled and pl_module.global_step > 0:
            self.log_img(pl_module, batch, batch_idx, split="val")
        if hasattr(pl_module, "calibrate_grad_norm"):
            if (pl_module.calibrate_grad_norm and batch_idx % 25 == 0) and batch_idx > 0:
                self.log_gradients(trainer, pl_module, batch_idx=batch_idx)


@rank_zero_only
def init_wandb(save_dir, opt, config, group_name, name_str):
    print(f"setting WANDB_DIR to {save_dir}")
    os.makedirs(save_dir, exist_ok=True)

    os.environ["WANDB_DIR"] = save_dir
    if opt.debug:
        wandb.init(project=opt.projectname, mode="offline", group=group_name)
    else:
        wandb.init(
            project=opt.projectname,
            config=config,
            settings=wandb.Settings(code_dir="./sgm"),
            group=group_name,
            name=name_str,
        )