path: root/data/csv.py

import math
import torch
import json
from functools import partial
from pathlib import Path
from typing import NamedTuple, Optional, Union, Callable

from PIL import Image

from torch.utils.data import IterableDataset, DataLoader, random_split
from torchvision import transforms
from transformers import CLIPTokenizer

from data.keywords import prompt_to_keywords, keywords_to_prompt
from models.clip.util import unify_input_ids


cache = {}


interpolations = {
    "linear": transforms.InterpolationMode.NEAREST,
    "bilinear": transforms.InterpolationMode.BILINEAR,
    "bicubic": transforms.InterpolationMode.BICUBIC,
    "lanczos": transforms.InterpolationMode.LANCZOS,
}


def get_image(path):
    if path in cache:
        return cache[path]

    image = Image.open(path)
    if not image.mode == "RGB":
        image = image.convert("RGB")
    cache[path] = image
    return image


def prepare_prompt(prompt: Union[str, dict[str, str]]):
    return {"content": prompt} if isinstance(prompt, str) else prompt


def generate_buckets(
    items: Union[list[str], list[Path]],
    base_size: int,
    step_size: int = 64,
    max_pixels: Optional[int] = None,
    num_buckets: int = 4,
    progressive_buckets: bool = False,
    return_tensor: bool = True
):
    if max_pixels is None:
        max_pixels = (base_size + step_size) ** 2

    max_pixels = max(max_pixels, base_size * base_size)

    bucket_items: list[int] = []
    bucket_assignments: list[int] = []
    buckets = [1.0]

    for i in range(1, num_buckets + 1):
        long_side = base_size + i * step_size
        short_side = min(base_size - math.ceil((base_size - max_pixels / long_side) / step_size) * step_size, base_size)
        buckets.append(long_side / short_side)
        buckets.append(short_side / long_side)

    buckets = torch.tensor(buckets)
    bucket_indices = torch.arange(len(buckets))

    for i, item in enumerate(items):
        image = get_image(item)
        ratio = image.width / image.height

        if ratio >= 1:
            mask = torch.logical_and(buckets >= 1, buckets <= ratio)
        else:
            mask = torch.logical_and(buckets <= 1, buckets >= ratio)

        if not progressive_buckets:
            inf = torch.zeros_like(buckets)
            inf[~mask] = math.inf
            mask = (buckets + inf - ratio).abs().argmin()

        indices = bucket_indices[mask]

        if len(indices.shape) == 0:
            indices = indices.unsqueeze(0)

        bucket_items += [i] * len(indices)
        bucket_assignments += indices

    if return_tensor:
        bucket_items = torch.tensor(bucket_items)
        bucket_assignments = torch.tensor(bucket_assignments)
    else:
        buckets = buckets.tolist()

    return buckets, bucket_items, bucket_assignments


def collate_fn(dtype: torch.dtype, tokenizer: CLIPTokenizer, with_prior_preservation: bool, examples):
    prompt_ids = [example["prompt_ids"] for example in examples]
    nprompt_ids = [example["nprompt_ids"] for example in examples]

    input_ids = [example["instance_prompt_ids"] for example in examples]
    pixel_values = [example["instance_images"] for example in examples]

    if with_prior_preservation:
        input_ids += [example["class_prompt_ids"] for example in examples]
        pixel_values += [example["class_images"] for example in examples]

    pixel_values = torch.stack(pixel_values)
    pixel_values = pixel_values.to(dtype=dtype, memory_format=torch.contiguous_format)

    prompts = unify_input_ids(tokenizer, prompt_ids)
    nprompts = unify_input_ids(tokenizer, nprompt_ids)
    inputs = unify_input_ids(tokenizer, input_ids)

    batch = {
        "prompt_ids": prompts.input_ids,
        "nprompt_ids": nprompts.input_ids,
        "input_ids": inputs.input_ids,
        "pixel_values": pixel_values,
        "attention_mask": inputs.attention_mask,
    }

    return batch


class VlpnDataItem(NamedTuple):
    instance_image_path: Path
    class_image_path: Path
    prompt: list[str]
    cprompt: str
    nprompt: str
    collection: list[str]


def keyword_filter(
    placeholder_tokens: Optional[list[str]],
    collection: Optional[list[str]],
    exclude_collections: Optional[list[str]],
    item: VlpnDataItem
):
    cond1 = placeholder_tokens is None or any(
        keyword in part
        for keyword in placeholder_tokens
        for part in item.prompt
    )
    cond2 = collection is None or collection in item.collection
    cond3 = exclude_collections is None or not any(
        collection in item.collection
        for collection in exclude_collections
    )
    return cond1 and cond2 and cond3


class VlpnDataModule():
    def __init__(
        self,
        batch_size: int,
        data_file: str,
        tokenizer: CLIPTokenizer,
        class_subdir: str = "cls",
        num_class_images: int = 1,
        size: int = 768,
        num_buckets: int = 0,
        bucket_step_size: int = 64,
        bucket_max_pixels: Optional[int] = None,
        progressive_buckets: bool = False,
        dropout: float = 0,
        shuffle: bool = False,
        interpolation: str = "bicubic",
        template_key: str = "template",
        valid_set_size: Optional[int] = None,
        train_set_pad: Optional[int] = None,
        valid_set_pad: Optional[int] = None,
        seed: Optional[int] = None,
        filter: Optional[Callable[[VlpnDataItem], bool]] = None,
        dtype: torch.dtype = torch.float32,
    ):
        super().__init__()

        self.data_file = Path(data_file)

        if not self.data_file.is_file():
            raise ValueError("data_file must be a file")

        self.data_root = self.data_file.parent
        self.class_root = self.data_root / class_subdir
        self.class_root.mkdir(parents=True, exist_ok=True)
        self.num_class_images = num_class_images

        self.tokenizer = tokenizer
        self.size = size
        self.num_buckets = num_buckets
        self.bucket_step_size = bucket_step_size
        self.bucket_max_pixels = bucket_max_pixels
        self.progressive_buckets = progressive_buckets
        self.dropout = dropout
        self.shuffle = shuffle
        self.template_key = template_key
        self.interpolation = interpolation
        self.valid_set_size = valid_set_size
        self.train_set_pad = train_set_pad if train_set_pad is not None else batch_size
        self.valid_set_pad = valid_set_pad if valid_set_pad is not None else batch_size
        self.seed = seed
        self.filter = filter
        self.batch_size = batch_size
        self.dtype = dtype

    def prepare_items(self, template, expansions, data) -> list[VlpnDataItem]:
        image = template["image"] if "image" in template else "{}"
        prompt = template["prompt"] if "prompt" in template else "{content}"
        cprompt = template["cprompt"] if "cprompt" in template else "{content}"
        nprompt = template["nprompt"] if "nprompt" in template else "{content}"

        return [
            VlpnDataItem(
                self.data_root / image.format(item["image"]),
                None,
                prompt_to_keywords(
                    prompt.format(**prepare_prompt(item["prompt"] if "prompt" in item else "")),
                    expansions
                ),
                keywords_to_prompt(prompt_to_keywords(
                    cprompt.format(**prepare_prompt(item["prompt"] if "prompt" in item else "")),
                    expansions
                )),
                keywords_to_prompt(prompt_to_keywords(
                    nprompt.format(**prepare_prompt(item["nprompt"] if "nprompt" in item else "")),
                    expansions
                )),
                item["collection"].split(", ") if "collection" in item else []
            )
            for item in data
        ]

    def filter_items(self, items: list[VlpnDataItem]) -> list[VlpnDataItem]:
        if self.filter is None:
            return items

        return [item for item in items if self.filter(item)]

    def pad_items(self, items: list[VlpnDataItem], num_class_images: int = 1) -> list[VlpnDataItem]:
        image_multiplier = max(num_class_images, 1)

        return [
            VlpnDataItem(
                item.instance_image_path,
                self.class_root / f"{item.instance_image_path.stem}_{i}{item.instance_image_path.suffix}",
                item.prompt,
                item.cprompt,
                item.nprompt,
                item.collection,
            )
            for item in items
            for i in range(image_multiplier)
        ]

    def setup(self):
        with open(self.data_file, 'rt') as f:
            metadata = json.load(f)
        template = metadata[self.template_key] if self.template_key in metadata else {}
        expansions = metadata["expansions"] if "expansions" in metadata else {}
        items = metadata["items"] if "items" in metadata else []

        items = self.prepare_items(template, expansions, items)
        items = self.filter_items(items)

        num_images = len(items)

        valid_set_size = min(self.valid_set_size, num_images) if self.valid_set_size is not None else num_images // 10
        train_set_size = max(num_images - valid_set_size, 1)
        valid_set_size = num_images - train_set_size

        generator = torch.Generator(device="cpu")
        if self.seed is not None:
            generator = generator.manual_seed(self.seed)

        collate_fn_ = partial(collate_fn, self.dtype, self.tokenizer, self.num_class_images != 0)

        if valid_set_size == 0:
            data_train, data_val = items, items[:self.batch_size]
        else:
            data_train, data_val = random_split(items, [train_set_size, valid_set_size], generator=generator)

        data_train = self.pad_items(data_train, self.num_class_images)

        if len(data_train) < self.train_set_pad:
            data_train *= math.ceil(self.train_set_pad / len(data_train))

        self.train_dataset = VlpnDataset(
            data_train, self.tokenizer,
            num_buckets=self.num_buckets, progressive_buckets=self.progressive_buckets,
            bucket_step_size=self.bucket_step_size, bucket_max_pixels=self.bucket_max_pixels,
            batch_size=self.batch_size, fill_batch=True, generator=generator,
            size=self.size, interpolation=self.interpolation,
            num_class_images=self.num_class_images, dropout=self.dropout, shuffle=self.shuffle,
        )

        self.train_dataloader = DataLoader(
            self.train_dataset,
            batch_size=None, pin_memory=True, collate_fn=collate_fn_
        )

        if len(data_val) != 0:
            data_val = self.pad_items(data_val)

            if len(data_val) < self.valid_set_pad:
                data_val *= math.ceil(self.valid_set_pad / len(data_val))

            self.val_dataset = VlpnDataset(
                data_val, self.tokenizer,
                num_buckets=self.num_buckets, progressive_buckets=True,
                bucket_step_size=self.bucket_step_size, bucket_max_pixels=self.bucket_max_pixels,
                batch_size=self.batch_size, generator=generator,
                size=self.size, interpolation=self.interpolation,
            )

            self.val_dataloader = DataLoader(
                self.val_dataset,
                batch_size=None, pin_memory=True, collate_fn=collate_fn_
            )
        else:
            self.val_dataloader = None


class VlpnDataset(IterableDataset):
    def __init__(
        self,
        items: list[VlpnDataItem],
        tokenizer: CLIPTokenizer,
        num_buckets: int = 1,
        bucket_step_size: int = 64,
        bucket_max_pixels: Optional[int] = None,
        progressive_buckets: bool = False,
        batch_size: int = 1,
        fill_batch: bool = False,
        num_class_images: int = 0,
        size: int = 768,
        dropout: float = 0,
        shuffle: bool = False,
        interpolation: str = "bicubic",
        generator: Optional[torch.Generator] = None,
    ):
        self.items = items
        self.batch_size = batch_size
        self.fill_batch = fill_batch

        self.tokenizer = tokenizer
        self.num_class_images = num_class_images
        self.size = size
        self.dropout = dropout
        self.shuffle = shuffle
        self.interpolation = interpolations[interpolation]
        self.generator = generator

        self.buckets, self.bucket_items, self.bucket_assignments = generate_buckets(
            [item.instance_image_path for item in self.items],
            base_size=size,
            step_size=bucket_step_size,
            num_buckets=num_buckets,
            max_pixels=bucket_max_pixels,
            progressive_buckets=progressive_buckets,
        )

        self.bucket_item_range = torch.arange(len(self.bucket_items))

        self.length_ = (self.bucket_assignments.bincount() / self.batch_size).ceil().long().sum().item()

    def get_input_ids(self, text: str):
        return self.tokenizer(text, padding="do_not_pad").input_ids

    def __len__(self):
        return self.length_

    def __iter__(self):
        worker_info = torch.utils.data.get_worker_info()

        if self.shuffle:
            perm = torch.randperm(len(self.bucket_assignments), generator=self.generator)
            self.bucket_items = self.bucket_items[perm]
            self.bucket_assignments = self.bucket_assignments[perm]

        image_transforms = None

        mask = torch.ones_like(self.bucket_assignments, dtype=bool)
        bucket = -1
        batch = []
        batch_size = self.batch_size

        if worker_info is not None:
            batch_size = math.ceil(batch_size / worker_info.num_workers)
            worker_batch = math.ceil(len(self) / worker_info.num_workers)
            start = worker_info.id * worker_batch
            end = start + worker_batch
            mask[:start] = False
            mask[end:] = False

        while mask.any() or len(batch) != 0:
            if len(batch) >= batch_size:
                yield batch
                batch = []
                continue

            bucket_mask = mask.logical_and(self.bucket_assignments == bucket)
            bucket_items = self.bucket_items[bucket_mask]

            if len(bucket_items) == 0 and len(batch) != 0 and not self.fill_batch:
                yield batch
                batch = []
                continue

            if len(bucket_items) == 0 and len(batch) == 0:
                bucket = self.bucket_assignments[mask][0]
                ratio = self.buckets[bucket]
                width = int(self.size * ratio) if ratio > 1 else self.size
                height = int(self.size / ratio) if ratio < 1 else self.size

                image_transforms = transforms.Compose(
                    [
                        transforms.Resize(self.size, interpolation=self.interpolation),
                        transforms.RandomCrop((height, width)),
                        transforms.RandomHorizontalFlip(),
                        transforms.ToTensor(),
                        transforms.Normalize([0.5], [0.5]),
                    ]
                )

                continue

            if len(bucket_items) == 0:
                bucket_items = self.bucket_items[self.bucket_assignments == bucket]
                item_index = bucket_items[torch.randint(len(bucket_items), (1,), generator=self.generator)]
            else:
                item_index = bucket_items[0]
                mask[self.bucket_item_range[bucket_mask][0]] = False

            item = self.items[item_index]

            example = {}

            example["prompt_ids"] = self.get_input_ids(keywords_to_prompt(item.prompt))
            example["nprompt_ids"] = self.get_input_ids(item.nprompt)

            example["instance_prompt_ids"] = self.get_input_ids(
                keywords_to_prompt(item.prompt, self.dropout, True)
            )
            example["instance_images"] = image_transforms(get_image(item.instance_image_path))

            if self.num_class_images != 0:
                example["class_prompt_ids"] = self.get_input_ids(item.cprompt)
                example["class_images"] = image_transforms(get_image(item.class_image_path))

            batch.append(example)