2022-10-02 15:03:39 +03:00
|
|
|
import os
|
|
|
|
import numpy as np
|
|
|
|
import PIL
|
|
|
|
import torch
|
|
|
|
from PIL import Image
|
2022-11-20 06:35:26 +03:00
|
|
|
from torch.utils.data import Dataset, DataLoader
|
2022-10-02 15:03:39 +03:00
|
|
|
from torchvision import transforms
|
|
|
|
|
|
|
|
import random
|
|
|
|
import tqdm
|
2022-10-11 19:03:08 +03:00
|
|
|
from modules import devices, shared
|
2022-10-04 08:52:11 +03:00
|
|
|
import re
|
|
|
|
|
2022-11-20 06:35:26 +03:00
|
|
|
from ldm.modules.distributions.distributions import DiagonalGaussianDistribution
|
|
|
|
|
2022-10-12 20:49:47 +03:00
|
|
|
re_numbers_at_start = re.compile(r"^[-\d]+\s*")
|
|
|
|
|
|
|
|
|
|
|
|
class DatasetEntry:
|
2022-11-20 06:35:26 +03:00
|
|
|
def __init__(self, filename=None, filename_text=None, latent_dist=None, latent_sample=None, cond=None, cond_text=None, pixel_values=None):
|
2022-10-12 20:49:47 +03:00
|
|
|
self.filename = filename
|
|
|
|
self.filename_text = filename_text
|
2022-11-20 06:35:26 +03:00
|
|
|
self.latent_dist = latent_dist
|
|
|
|
self.latent_sample = latent_sample
|
|
|
|
self.cond = cond
|
|
|
|
self.cond_text = cond_text
|
|
|
|
self.pixel_values = pixel_values
|
2022-10-02 15:03:39 +03:00
|
|
|
|
|
|
|
|
|
|
|
class PersonalizedBase(Dataset):
|
2022-11-20 06:35:26 +03:00
|
|
|
def __init__(self, data_root, width, height, repeats, flip_p=0.5, placeholder_token="*", model=None, cond_model=None, device=None, template_file=None, include_cond=False, batch_size=1, gradient_step=1, shuffle_tags=False, tag_drop_out=0, latent_sampling_method='once'):
|
2022-10-15 09:24:59 +03:00
|
|
|
re_word = re.compile(shared.opts.dataset_filename_word_regex) if len(shared.opts.dataset_filename_word_regex) > 0 else None
|
2022-11-20 06:35:26 +03:00
|
|
|
|
2022-10-02 15:03:39 +03:00
|
|
|
self.placeholder_token = placeholder_token
|
|
|
|
|
2022-10-10 16:35:35 +03:00
|
|
|
self.width = width
|
|
|
|
self.height = height
|
2022-10-02 15:03:39 +03:00
|
|
|
self.flip = transforms.RandomHorizontalFlip(p=flip_p)
|
|
|
|
|
|
|
|
self.dataset = []
|
|
|
|
|
|
|
|
with open(template_file, "r") as file:
|
|
|
|
lines = [x.strip() for x in file.readlines()]
|
|
|
|
|
|
|
|
self.lines = lines
|
|
|
|
|
|
|
|
assert data_root, 'dataset directory not specified'
|
2022-10-29 15:44:05 +03:00
|
|
|
assert os.path.isdir(data_root), "Dataset directory doesn't exist"
|
|
|
|
assert os.listdir(data_root), "Dataset directory is empty"
|
2022-10-02 15:03:39 +03:00
|
|
|
|
2022-10-11 11:32:46 +03:00
|
|
|
self.image_paths = [os.path.join(data_root, file_path) for file_path in os.listdir(data_root)]
|
2022-11-20 06:35:26 +03:00
|
|
|
|
|
|
|
|
|
|
|
self.shuffle_tags = shuffle_tags
|
|
|
|
self.tag_drop_out = tag_drop_out
|
|
|
|
|
2022-10-02 15:03:39 +03:00
|
|
|
print("Preparing dataset...")
|
|
|
|
for path in tqdm.tqdm(self.image_paths):
|
2022-11-20 06:35:26 +03:00
|
|
|
if shared.state.interrupted:
|
|
|
|
raise Exception("inturrupted")
|
2022-10-11 11:32:46 +03:00
|
|
|
try:
|
|
|
|
image = Image.open(path).convert('RGB').resize((self.width, self.height), PIL.Image.BICUBIC)
|
|
|
|
except Exception:
|
|
|
|
continue
|
2022-10-02 15:03:39 +03:00
|
|
|
|
2022-10-12 20:49:47 +03:00
|
|
|
text_filename = os.path.splitext(path)[0] + ".txt"
|
2022-10-02 15:03:39 +03:00
|
|
|
filename = os.path.basename(path)
|
2022-10-12 20:49:47 +03:00
|
|
|
|
|
|
|
if os.path.exists(text_filename):
|
|
|
|
with open(text_filename, "r", encoding="utf8") as file:
|
|
|
|
filename_text = file.read()
|
|
|
|
else:
|
|
|
|
filename_text = os.path.splitext(filename)[0]
|
|
|
|
filename_text = re.sub(re_numbers_at_start, '', filename_text)
|
|
|
|
if re_word:
|
|
|
|
tokens = re_word.findall(filename_text)
|
|
|
|
filename_text = (shared.opts.dataset_filename_join_string or "").join(tokens)
|
2022-10-02 15:03:39 +03:00
|
|
|
|
|
|
|
npimage = np.array(image).astype(np.uint8)
|
|
|
|
npimage = (npimage / 127.5 - 1.0).astype(np.float32)
|
|
|
|
|
2022-11-20 06:35:26 +03:00
|
|
|
torchdata = torch.from_numpy(npimage).permute(2, 0, 1).to(device=device, dtype=torch.float32)
|
|
|
|
latent_sample = None
|
|
|
|
|
2022-11-29 05:36:35 +03:00
|
|
|
with devices.autocast():
|
2022-11-20 06:35:26 +03:00
|
|
|
latent_dist = model.encode_first_stage(torchdata.unsqueeze(dim=0))
|
|
|
|
|
|
|
|
if latent_sampling_method == "once" or (latent_sampling_method == "deterministic" and not isinstance(latent_dist, DiagonalGaussianDistribution)):
|
|
|
|
latent_sample = model.get_first_stage_encoding(latent_dist).squeeze().to(devices.cpu)
|
|
|
|
latent_sampling_method = "once"
|
|
|
|
entry = DatasetEntry(filename=path, filename_text=filename_text, latent_sample=latent_sample)
|
|
|
|
elif latent_sampling_method == "deterministic":
|
|
|
|
# Works only for DiagonalGaussianDistribution
|
|
|
|
latent_dist.std = 0
|
|
|
|
latent_sample = model.get_first_stage_encoding(latent_dist).squeeze().to(devices.cpu)
|
|
|
|
entry = DatasetEntry(filename=path, filename_text=filename_text, latent_sample=latent_sample)
|
|
|
|
elif latent_sampling_method == "random":
|
|
|
|
entry = DatasetEntry(filename=path, filename_text=filename_text, latent_dist=latent_dist)
|
|
|
|
|
|
|
|
if not (self.tag_drop_out != 0 or self.shuffle_tags):
|
2022-10-12 20:49:47 +03:00
|
|
|
entry.cond_text = self.create_text(filename_text)
|
2022-10-11 19:03:08 +03:00
|
|
|
|
2022-11-20 06:35:26 +03:00
|
|
|
if include_cond and not (self.tag_drop_out != 0 or self.shuffle_tags):
|
2022-11-29 05:36:35 +03:00
|
|
|
with devices.autocast():
|
2022-11-20 06:35:26 +03:00
|
|
|
entry.cond = cond_model([entry.cond_text]).to(devices.cpu).squeeze(0)
|
2022-10-02 15:03:39 +03:00
|
|
|
|
2022-11-20 06:35:26 +03:00
|
|
|
self.dataset.append(entry)
|
|
|
|
del torchdata
|
|
|
|
del latent_dist
|
|
|
|
del latent_sample
|
2022-10-02 15:03:39 +03:00
|
|
|
|
2022-11-20 06:35:26 +03:00
|
|
|
self.length = len(self.dataset)
|
|
|
|
assert self.length > 0, "No images have been found in the dataset."
|
|
|
|
self.batch_size = min(batch_size, self.length)
|
|
|
|
self.gradient_step = min(gradient_step, self.length // self.batch_size)
|
|
|
|
self.latent_sampling_method = latent_sampling_method
|
2022-10-02 15:03:39 +03:00
|
|
|
|
2022-10-12 20:49:47 +03:00
|
|
|
def create_text(self, filename_text):
|
2022-10-11 19:03:08 +03:00
|
|
|
text = random.choice(self.lines)
|
2022-11-11 04:29:55 +03:00
|
|
|
tags = filename_text.split(',')
|
2022-11-20 06:35:26 +03:00
|
|
|
if self.tag_drop_out != 0:
|
|
|
|
tags = [t for t in tags if random.random() > self.tag_drop_out]
|
|
|
|
if self.shuffle_tags:
|
2022-11-01 17:29:12 +03:00
|
|
|
random.shuffle(tags)
|
2022-11-11 04:29:55 +03:00
|
|
|
text = text.replace("[filewords]", ','.join(tags))
|
2022-11-13 22:44:27 +03:00
|
|
|
text = text.replace("[name]", self.placeholder_token)
|
2022-10-11 19:03:08 +03:00
|
|
|
return text
|
|
|
|
|
2022-10-02 15:03:39 +03:00
|
|
|
def __len__(self):
|
|
|
|
return self.length
|
|
|
|
|
|
|
|
def __getitem__(self, i):
|
2022-11-20 06:35:26 +03:00
|
|
|
entry = self.dataset[i]
|
|
|
|
if self.tag_drop_out != 0 or self.shuffle_tags:
|
|
|
|
entry.cond_text = self.create_text(entry.filename_text)
|
|
|
|
if self.latent_sampling_method == "random":
|
2022-11-20 10:14:27 +03:00
|
|
|
entry.latent_sample = shared.sd_model.get_first_stage_encoding(entry.latent_dist).to(devices.cpu)
|
2022-11-20 06:35:26 +03:00
|
|
|
return entry
|
|
|
|
|
|
|
|
class PersonalizedDataLoader(DataLoader):
|
2022-11-21 04:15:46 +03:00
|
|
|
def __init__(self, dataset, latent_sampling_method="once", batch_size=1, pin_memory=False):
|
|
|
|
super(PersonalizedDataLoader, self).__init__(dataset, shuffle=True, drop_last=True, batch_size=batch_size, pin_memory=pin_memory)
|
|
|
|
if latent_sampling_method == "random":
|
|
|
|
self.collate_fn = collate_wrapper_random
|
|
|
|
else:
|
|
|
|
self.collate_fn = collate_wrapper
|
2022-11-20 06:35:26 +03:00
|
|
|
|
|
|
|
|
|
|
|
class BatchLoader:
|
|
|
|
def __init__(self, data):
|
|
|
|
self.cond_text = [entry.cond_text for entry in data]
|
|
|
|
self.cond = [entry.cond for entry in data]
|
|
|
|
self.latent_sample = torch.stack([entry.latent_sample for entry in data]).squeeze(1)
|
2022-11-21 04:15:46 +03:00
|
|
|
#self.emb_index = [entry.emb_index for entry in data]
|
|
|
|
#print(self.latent_sample.device)
|
2022-11-20 06:35:26 +03:00
|
|
|
|
|
|
|
def pin_memory(self):
|
|
|
|
self.latent_sample = self.latent_sample.pin_memory()
|
|
|
|
return self
|
|
|
|
|
|
|
|
def collate_wrapper(batch):
|
2022-11-21 04:15:46 +03:00
|
|
|
return BatchLoader(batch)
|
|
|
|
|
|
|
|
class BatchLoaderRandom(BatchLoader):
|
|
|
|
def __init__(self, data):
|
|
|
|
super().__init__(data)
|
|
|
|
|
|
|
|
def pin_memory(self):
|
|
|
|
return self
|
|
|
|
|
|
|
|
def collate_wrapper_random(batch):
|
|
|
|
return BatchLoaderRandom(batch)
|