open-magvit2 1.1

Packaging of Open-source replication of Google's MAGVIT-v2 tokenizer

OPEN-MAGVIT2 package

This is a fork of OPEN-MAGVIT2: An Open-source Project Toward Democratizing Auto-Regressive Visual Generation in order to make it a package for easy usage.

Install

pip install open-magvit2

Example of usage

1. Download the checkpoint from huggingface

wget https://huggingface.co/TencentARC/Open-MAGVIT2/resolve/main/imagenet_256_L.ckpt

2. Load the model

import pkg_resources
import torch
from omegaconf import OmegaConf
from open_magvit2.reconstruct import load_vqgan_new

DEVICE = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
config_path = pkg_resources.resource_filename('open_magvit2', 'configs/gpu/imagenet_lfqgan_256_L.yaml')
config = OmegaConf.load(config_path)
model = load_vqgan_new(config, "imagenet_256_L.ckpt").to(DEVICE)

3. Encode an image

from PIL import Image
import torchvision.transforms as transforms

image = Image.open('1165.jpg')
image_tensor = transforms.ToTensor()(image)
batch = image_tensor.unsqueeze(0)
with torch.no_grad():
  quant, emb_loss, tokens, loss_breakdown = model.encode(image_tensor)

4. Decode

• decode from embeddings

from open_magvit2.reconstruct import custom_to_pil

with torch.no_grad():
    tensor = model.decode(quant)

reconstructed_image = custom_to_pil(tensor[0])

• decode from tokens (i.e. ids)

from einops import rearrange
from open_magvit2.reconstruct import custom_to_pil

x = rearrange(tokens, "(b s) -> b s", b=1)
q = model.quantize.get_codebook_entry(x, (1, 16, 16, 18), order='')

with torch.no_grad():
    tensor2 = model.decode(q)

reconstructed_image2 = custom_to_pil(tensor2[0])

Example colab

Check this notebook open-MAGVIT2-package-inference-example.ipynb