app.py

#!/usr/bin/env python

from __future__ import annotations

import argparse
import functools
import os
import pickle
import sys

import gradio as gr
import numpy as np
import torch
import torch.nn as nn
from huggingface_hub import hf_hub_download

sys.path.insert(0, 'stylegan3')

TITLE = 'StyleGAN2'
DESCRIPTION = '''This is an unofficial demo for https://github.com/NVlabs/stylegan3.

Expected execution time on Hugging Face Spaces: 4s
'''
SAMPLE_IMAGE_DIR = 'https://huggingface.co/spaces/hysts/StyleGAN2/resolve/main/samples'
ARTICLE = f'''## Generated images
- truncation: 0.7
### CIFAR-10
- size: 32x32
- class index: 0-9
- seed: 0-9
![CIFAR-10 samples]({SAMPLE_IMAGE_DIR}/cifar10.jpg)
### AFHQ-Cat
- size: 512x512
- seed: 0-99
![AFHQ-Cat samples]({SAMPLE_IMAGE_DIR}/afhq-cat.jpg)
### AFHQ-Dog
- size: 512x512
- seed: 0-99
![AFHQ-Dog samples]({SAMPLE_IMAGE_DIR}/afhq-dog.jpg)
### AFHQ-Wild
- size: 512x512
- seed: 0-99
![AFHQ-Wild samples]({SAMPLE_IMAGE_DIR}/afhq-wild.jpg)
### AFHQv2
- size: 512x512
- seed: 0-99
![AFHQv2 samples]({SAMPLE_IMAGE_DIR}/afhqv2.jpg)
### LSUN-Dog
- size: 256x256
- seed: 0-99
![LSUN-Dog samples]({SAMPLE_IMAGE_DIR}/lsun-dog.jpg)
### BreCaHAD
- size: 512x512
- seed: 0-99
![BreCaHAD samples]({SAMPLE_IMAGE_DIR}/brecahad.jpg)
### CelebA-HQ
- size: 256x256
- seed: 0-99
![CelebA-HQ samples]({SAMPLE_IMAGE_DIR}/celebahq.jpg)
### FFHQ
- size: 1024x1024
- seed: 0-99
![FFHQ samples]({SAMPLE_IMAGE_DIR}/ffhq.jpg)
### FFHQ-U
- size: 1024x1024
- seed: 0-99
![FFHQ-U samples]({SAMPLE_IMAGE_DIR}/ffhq-u.jpg)
### MetFaces
- size: 1024x1024
- seed: 0-99
![MetFaces samples]({SAMPLE_IMAGE_DIR}/metfaces.jpg)
### MetFaces-U
- size: 1024x1024
- seed: 0-99
![MetFaces-U samples]({SAMPLE_IMAGE_DIR}/metfaces-u.jpg)

<center><img src="https://visitor-badge.glitch.me/badge?page_id=hysts.stylegan2" alt="visitor badge"/></center>
'''

TOKEN = os.environ['TOKEN']


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser()
    parser.add_argument('--device', type=str, default='cpu')
    parser.add_argument('--theme', type=str)
    parser.add_argument('--live', action='store_true')
    parser.add_argument('--share', action='store_true')
    parser.add_argument('--port', type=int)
    parser.add_argument('--disable-queue',
                        dest='enable_queue',
                        action='store_false')
    parser.add_argument('--allow-flagging', type=str, default='never')
    return parser.parse_args()


def generate_z(z_dim: int, seed: int, device: torch.device) -> torch.Tensor:
    return torch.from_numpy(np.random.RandomState(seed).randn(
        1, z_dim)).to(device).float()


@torch.inference_mode()
def generate_image(model_name: str, class_index: int, seed: int,
                   truncation_psi: float, model_dict: dict[str, nn.Module],
                   device: torch.device) -> np.ndarray:
    model = model_dict[model_name]
    seed = int(np.clip(seed, 0, np.iinfo(np.uint32).max))

    z = generate_z(model.z_dim, seed, device)
    label = torch.zeros([1, model.c_dim], device=device)
    class_index = round(class_index)
    class_index = min(max(0, class_index), model.c_dim - 1)
    class_index = torch.tensor(class_index, dtype=torch.long)
    if class_index >= 0:
        label[:, class_index] = 1

    out = model(z, label, truncation_psi=truncation_psi)
    out = (out.permute(0, 2, 3, 1) * 127.5 + 128).clamp(0, 255).to(torch.uint8)
    return out[0].cpu().numpy()


def load_model(file_name: str, device: torch.device) -> nn.Module:
    path = hf_hub_download('hysts/StyleGAN2',
                           f'models/{file_name}',
                           use_auth_token=TOKEN)
    with open(path, 'rb') as f:
        model = pickle.load(f)['G_ema']
    model.eval()
    model.to(device)
    with torch.inference_mode():
        z = torch.zeros((1, model.z_dim)).to(device)
        label = torch.zeros([1, model.c_dim], device=device)
        model(z, label)
    return model


def main():
    args = parse_args()
    device = torch.device(args.device)

    model_names = {
        'AFHQ-Cat-512': 'stylegan2-afhqcat-512x512.pkl',
        'AFHQ-Dog-512': 'stylegan2-afhqdog-512x512.pkl',
        'AFHQv2-512': 'stylegan2-afhqv2-512x512.pkl',
        'AFHQ-Wild-512': 'stylegan2-afhqwild-512x512.pkl',
        'BreCaHAD-512': 'stylegan2-brecahad-512x512.pkl',
        'CelebA-HQ-256': 'stylegan2-celebahq-256x256.pkl',
        'CIFAR-10': 'stylegan2-cifar10-32x32.pkl',
        'FFHQ-256': 'stylegan2-ffhq-256x256.pkl',
        'FFHQ-512': 'stylegan2-ffhq-512x512.pkl',
        'FFHQ-1024': 'stylegan2-ffhq-1024x1024.pkl',
        'FFHQ-U-256': 'stylegan2-ffhqu-256x256.pkl',
        'FFHQ-U-1024': 'stylegan2-ffhqu-1024x1024.pkl',
        'LSUN-Dog-256': 'stylegan2-lsundog-256x256.pkl',
        'MetFaces-1024': 'stylegan2-metfaces-1024x1024.pkl',
        'MetFaces-U-1024': 'stylegan2-metfacesu-1024x1024.pkl',
    }

    model_dict = {
        name: load_model(file_name, device)
        for name, file_name in model_names.items()
    }

    func = functools.partial(generate_image,
                             model_dict=model_dict,
                             device=device)
    func = functools.update_wrapper(func, generate_image)

    gr.Interface(
        func,
        [
            gr.inputs.Radio(list(model_names.keys()),
                            type='value',
                            default='FFHQ-1024',
                            label='Model'),
            gr.inputs.Number(default=0, label='Class index'),
            gr.inputs.Number(default=0, label='Seed'),
            gr.inputs.Slider(
                0, 2, step=0.05, default=0.7, label='Truncation psi'),
        ],
        gr.outputs.Image(type='numpy', label='Output'),
        title=TITLE,
        description=DESCRIPTION,
        article=ARTICLE,
        theme=args.theme,
        allow_flagging=args.allow_flagging,
        live=args.live,
    ).launch(
        enable_queue=args.enable_queue,
        server_port=args.port,
        share=args.share,
    )


if __name__ == '__main__':
    main()