app.py

from typing import Literal
import gradio as gr
import torch
import numpy as np
import colorsys
import yaml

from huggingface_hub import hf_hub_download
from diffusers import VQModel
from diffusers.image_processor import VaeImageProcessor
from diffusers.pipelines.wuerstchen.modeling_paella_vq_model import PaellaVQModel

from chameleon.image_tokenizer import ImageTokenizer

import torch.backends
import torch.mps
from PIL import Image

import spaces


Model = Literal["vqgan", "paella", "chameleon"]
models = ["vqgan", "paella", "chameleon"]

if torch.cuda.is_available():
    device = torch.device("cuda")
elif torch.backends.mps.is_available():
    device = torch.device("mps")
else:
    device = torch.device("cpu")


class ImageRoundtripPipeline:
    def roundtrip_image(self, image, output_type="pil"): ...


class VQImageRoundtripPipeline(ImageRoundtripPipeline):
    vqvae: VQModel
    vae_scale_factor: int
    vqvae_processor: VaeImageProcessor

    def __init__(self):
        self.vqvae = VQModel.from_pretrained("amused/amused-512", subfolder="vqvae")
        self.vqvae.eval()
        self.vqvae.to(device)
        self.vae_scale_factor = 2 ** (len(self.vqvae.config.block_out_channels) - 1)
        self.vqvae_processor = VaeImageProcessor(
            vae_scale_factor=self.vae_scale_factor, do_normalize=False
        )
        print("VQ-GAN model loaded", self.vqvae)

    def roundtrip_image(self, image, output_type="pil"):
        image = self.vqvae_processor.preprocess(image)
        device = self.vqvae.device
        needs_upcasting = (
            self.vqvae.dtype == torch.float16 and self.vqvae.config.force_upcast
        )

        batch_size, im_channels, height, width = image.shape

        if needs_upcasting:
            self.vqvae.float()

        latents = self.vqvae.encode(
            image.to(dtype=self.vqvae.dtype, device=device)
        ).latents
        latents_batch_size, latent_channels, latents_height, latents_width = (
            latents.shape
        )
        latents = self.vqvae.quantize(latents)[2][2].reshape(
            batch_size, latents_height, latents_width
        )
        # replace 20% of latents with random values
        # random_latents = torch.randint(
        #     0, self.vqvae.config.num_vq_embeddings, latents.shape, device=device
        # )
        # random_mask = torch.rand(latents.shape, device=device) < 0.2
        # latents = torch.where(random_mask, random_latents, latents)
        output = self.vqvae.decode(
            latents,
            force_not_quantize=True,
            shape=(
                batch_size,
                height // self.vae_scale_factor,
                width // self.vae_scale_factor,
                self.vqvae.config.latent_channels,
            ),
        ).sample.clip(0, 1)
        output = self.vqvae_processor.postprocess(output, output_type)

        if needs_upcasting:
            self.vqvae.half()

        return output[0], latents.cpu().numpy(), self.vqvae.config.num_vq_embeddings


class ChameleonVQImageRoundtripPipeline(ImageRoundtripPipeline):
    tokenizer: ImageTokenizer
    n_embed: int
    vae_scale_factor: int

    def __init__(self):
        vqgan_path = hf_hub_download(
            "darknoon/chameleon-tokenizer", "tokenizer/vqgan.ckpt"
        )
        vqgan_config_path = hf_hub_download(
            "darknoon/chameleon-tokenizer", "tokenizer/vqgan.yaml"
        )
        self.tokenizer = ImageTokenizer(
            cfg_path=vqgan_config_path, ckpt_path=vqgan_path, device=device
        )
        with open(vqgan_config_path) as f:
            vq_config = yaml.safe_load(f)

        self.n_embed = vq_config["model"]["params"]["n_embed"]
        self.vae_scale_factor = 16
        print("Chameleon VQGan model loaded", self.tokenizer._vq_model, self.n_embed)

    def preprocess(self, image: Image):
        # copied from _vqgan_input_from
        np_img = np.array(image) / 255.0  # Normalize to [0, 1]
        np_img = np_img * 2 - 1  # Scale to [-1, 1]
        tensor_img = (
            torch.from_numpy(np_img).permute(2, 0, 1).float()
        )  # (Channels, Height, Width) format.

        # Add batch dimension.
        return tensor_img.unsqueeze(0)

    def roundtrip_image(self, image, output_type="pil"):
        # image = self.tokenizer._vqgan_input_from(image).to(device)
        image = self.preprocess(image).to(device)
        _, _, im_height, im_width = image.shape
        _, _, [_, _, latents] = self.tokenizer._vq_model.encode(image)
        scale = self.vae_scale_factor
        shape = (1, im_height // scale, im_width // scale)
        output = self.tokenizer.pil_from_img_toks(latents, shape=shape)
        # we actually do want this to be a grid, sorry!
        latents = latents.reshape(*shape)

        return (
            output,
            latents.cpu().numpy(),
            self.n_embed,
        )


class PaellaImageRoundtripPipeline(ImageRoundtripPipeline):
    vqgan: PaellaVQModel
    vae_scale_factor: int
    vqvae_processor: VaeImageProcessor

    def __init__(self):
        self.vqgan = PaellaVQModel.from_pretrained(
            "warp-ai/wuerstchen", subfolder="vqgan"
        )
        self.vqgan.eval()
        self.vqgan.to(device)
        self.vae_scale_factor = 4
        self.vqvae_processor = VaeImageProcessor(
            vae_scale_factor=self.vae_scale_factor, do_normalize=False
        )
        print("Paella VQ-GAN model loaded", self.vqgan)

    def roundtrip_image(self, image, output_type="pil"):
        image = self.vqvae_processor.preprocess(image)
        device = self.vqgan.device

        batch_size, im_channels, height, width = image.shape

        latents = self.vqgan.encode(
            image.to(dtype=self.vqgan.dtype, device=device)
        ).latents
        latents_batch_size, latent_channels, latents_height, latents_width = (
            latents.shape
        )
        # latents = latents * self.vqgan.config.scale_factor
        # Manually quantize so we can inspect
        latents_q = self.vqgan.vquantizer(latents)[2][2].reshape(
            batch_size, latents_height, latents_width
        )
        print("latents after quantize", (latents_q.shape, latents_q.dtype))
        images = self.vqgan.decode(latents).sample.clamp(0, 1)
        output = self.vqvae_processor.postprocess(images, output_type)

        # if needs_upcasting:
        #     self.vqgan.half()

        return output[0], latents_q.cpu().numpy(), self.vqgan.config.num_vq_embeddings


pipeline_paella = PaellaImageRoundtripPipeline()
pipeline_vq = VQImageRoundtripPipeline()
pipeline_vq_chameleon = ChameleonVQImageRoundtripPipeline()


# Function to generate a list of unique colors
def generate_unique_colors_hsl(n):
    colors = []
    for i in range(n):
        hue = i / (n // 4)  # Distribute hues evenly around the color wheel 4 times
        lightness = 0.8 - (i / n) * 0.6  # Decrease brightness from 0.8 to 0.2
        saturation = 1.0
        rgb = colorsys.hls_to_rgb(hue, lightness, saturation)
        rgb = tuple(int(255 * x) for x in rgb)
        colors.append(rgb)
    return colors


# Function to create the image from VQGAN tokens
def vqgan_tokens_to_image(tokens, codebook_size, downscale_factor):
    # Generate unique colors for each token in the codebook
    colors = generate_unique_colors_hsl(codebook_size)

    # Create a lookup table
    lookup_table = np.array(colors, dtype=np.uint8)

    # Extract the token array (remove the batch dimension)
    token_array = tokens[0]

    # Map tokens to their RGB colors using the lookup table
    color_image = lookup_table[token_array]

    # Create a PIL image from the numpy array
    img = Image.fromarray(color_image, "RGB")

    # Upscale the image using nearest neighbor interpolation
    img = img.resize(
        (
            color_image.shape[1] * downscale_factor,
            color_image.shape[0] * downscale_factor,
        ),
        Image.NEAREST,
    )

    return img


def describe_shape(shape):
    return f"Shape: {shape} num elements: {np.prod(shape)}"


def calc_psnr(img1: Image, img2: Image):
    if img1.size != img2.size:
        raise ValueError("Images must have the same dimensions")
    img1 = np.array(img1)
    img2 = np.array(img2)
    mse = np.mean((img1 - img2) ** 2)
    if mse == 0:
        return float("inf")
    return 2 * 10 * np.log10(255.0 / np.sqrt(mse))


@spaces.GPU(duration=32)
@torch.no_grad()
def roundtrip_image(
    image,
    model: Model,
    size: Literal["256x256", "512x512", "1024x1024"],
    output_type="pil",
):
    if size == "256x256":
        image = image.resize((256, 256))
    elif size == "512x512":
        image = image.resize((512, 512))
    elif size == "1024x1024":
        image = image.resize((1024, 1024))
    else:
        raise ValueError(f"Unknown size {size}")

    image_orig = image
    if model == "vqgan":
        pipeline = pipeline_vq
    elif model == "paella":
        pipeline = pipeline_paella
    elif model == "chameleon":
        pipeline = pipeline_vq_chameleon
    else:
        raise ValueError(f"Unknown model {model}")

    image, latents, codebook_size = pipeline.roundtrip_image(image, output_type)

    return (
        image,
        vqgan_tokens_to_image(
            latents, codebook_size, downscale_factor=pipeline.vae_scale_factor
        ),
        describe_shape(latents.shape),
        f"{calc_psnr(image_orig, image):.2f}",
    )


demo = gr.Interface(
    fn=roundtrip_image,
    inputs=[
        gr.Image(type="pil"),
        gr.Dropdown(models, label="Model", value="vqgan"),
        gr.Dropdown(["256x256", "512x512", "1024x1024"], label="Size", value="512x512"),
    ],
    outputs=[
        gr.Image(label="Reconstructed", format="png"),
        gr.Image(label="Tokens", format="png"),
        gr.Text(label="VQ Shape"),
        gr.Text(label="PSNR"),
    ],
    title="Image Tokenizer Playground",
    description="Round-trip an image through an encode-decoder pair to see the quality loss from the VQ-GAN for image generation, etc.",
)

demo.launch()