mirror of
https://github.com/finegrain-ai/refiners.git
synced 2024-11-09 23:12:02 +00:00
Make image resize configurable in training scripts
This commit is contained in:
parent
437fa24368
commit
08a5341452
|
@ -211,6 +211,8 @@ class HuggingfaceDatasetConfig(BaseModel):
|
|||
horizontal_flip: bool = False
|
||||
random_crop: bool = True
|
||||
use_verification: bool = False
|
||||
resize_image_min_size: int = 512
|
||||
resize_image_max_size: int = 576
|
||||
|
||||
|
||||
class CheckpointingConfig(BaseModel):
|
||||
|
|
|
@ -98,7 +98,11 @@ class TextEmbeddingLatentsDataset(Dataset[TextEmbeddingLatentsBatch]):
|
|||
def __getitem__(self, index: int) -> TextEmbeddingLatentsBatch:
|
||||
item = self.dataset[index]
|
||||
caption, image = item["caption"], item["image"]
|
||||
resized_image = self.resize_image(image=image)
|
||||
resized_image = self.resize_image(
|
||||
image=image,
|
||||
min_size=self.config.dataset.resize_image_min_size,
|
||||
max_size=self.config.dataset.resize_image_max_size,
|
||||
)
|
||||
processed_image = self.process_image(resized_image)
|
||||
latents = self.lda.encode_image(image=processed_image).to(device=self.device)
|
||||
processed_caption = self.process_caption(caption=caption)
|
||||
|
|
Loading…
Reference in a new issue