| from typing import OrderedDict |
| import os |
| import json |
| import torch |
| from torch.utils.data import Dataset |
| from backend.torch.utils import load_image |
|
|
| class LimitedCache: |
| def __init__(self, max_size): |
| self.cache = OrderedDict() |
| self.max_size = max_size |
|
|
| def get(self, key, loader_fn): |
| if key in self.cache: |
| self.cache.move_to_end(key) |
| return self.cache[key] |
| else: |
| value = loader_fn(key) |
| self.cache[key] = value |
| if len(self.cache) > self.max_size: |
| self.cache.popitem(last=False) |
| return value |
|
|
| class CaptionControlDataset(Dataset): |
| @staticmethod |
| def collate_fn(batch): |
| return [(prompt, image, control) for prompt, image, control in batch] |
|
|
| def __init__(self, path, cache_size=1024): |
| super().__init__() |
| self.base_path = path |
| with open(os.path.join(path, "metadata.json"), "r") as f: |
| self.metadata = json.load(f) |
|
|
| self.image_cache = LimitedCache(cache_size) |
| self.control_cache = LimitedCache(cache_size) |
|
|
| def __len__(self): |
| return len(self.metadata) |
|
|
| def __getitem__(self, idx): |
| item = self.metadata[idx] |
| prompt = item["prompt"] |
| image_path = os.path.join(self.base_path, item["image"]) |
| control_path = os.path.join(self.base_path, item["control"]) |
|
|
| image = self.image_cache.get(image_path, load_image) |
| control = self.control_cache.get(control_path, load_image) |
|
|
| return prompt, image, control |
|
|
| def get_dataloader(self, batch_size=1, shuffle=False, **kwargs): |
| return torch.utils.data.DataLoader( |
| self, |
| batch_size=batch_size, |
| shuffle=shuffle, |
| collate_fn=self.collate_fn, |
| **kwargs, |
| ) |
|
|
