metadata
dataset_info:
features:
- name: label
dtype:
class_label:
names:
'0': '0'
'1': '1'
'2': '2'
'3': '3'
'4': '4'
'5': '5'
'6': '6'
'7': '7'
'8': '8'
'9': '9'
'10': a
'11': b
'12': c
'13': d
'14': e
'15': f
- name: latent
sequence:
sequence:
sequence: float32
splits:
- name: test
num_bytes: 27646560
num_examples: 6312
- name: train
num_bytes: 525227700
num_examples: 119915
download_size: 527167710
dataset_size: 552874260
Dataset Card for "latent_lsun_church_128px"
Each image is cropped to 128px square and encoded to a 4x16x16 latent representation using the same VAE as that employed by Stable Diffusion
Decoding
from diffusers import AutoencoderKL
from datasets import load_dataset
from PIL import Image
import numpy as np
import torch
# load the dataset
dataset = load_dataset('tglcourse/latent_lsun_church_128px')
# Load the VAE (requires access - see repo model card for info)
vae = AutoencoderKL.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="vae")
latent = torch.tensor([dataset['train'][0]['latent']]) # To tensor (bs, 4, 16, 16)
latent = (1 / 0.18215) * latent # Scale to match SD implementation
with torch.no_grad():
image = vae.decode(latent).sample[0] # Decode
image = (image / 2 + 0.5).clamp(0, 1) # To (0, 1)
image = image.detach().cpu().permute(1, 2, 0).numpy() # To numpy, channels lsat
image = (image * 255).round().astype("uint8") # (0, 255) and type uint8
image = Image.fromarray(image) # To PIL
image # The resulting PIL image