app.py

#!/usr/bin/env python

from __future__ import annotations

import argparse
import functools
import io
import os
import pathlib
import tarfile

import gradio as gr
import numpy as np
import PIL.Image
from huggingface_hub import hf_hub_download

TITLE = 'TADNE (This Anime Does Not Exist) Image Selector'
DESCRIPTION = '''The original TADNE site is https://thisanimedoesnotexist.ai/.

You can view images generated by the TADNE model with seed 0-99999.
You can filter images based on predictions by the [DeepDanbooru](https://github.com/KichangKim/DeepDanbooru) model.
The original images are 512x512 in size, but here they are resized to 128x128.

Known issues:
- The `Seed` table in the output doesn't refresh properly in gradio 2.9.1. https://github.com/gradio-app/gradio/issues/921
'''
ARTICLE = None

TOKEN = os.environ['TOKEN']


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser()
    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')
    parser.add_argument('--allow-screenshot', action='store_true')
    return parser.parse_args()


def download_image_tarball(size: int, dirname: str) -> pathlib.Path:
    path = hf_hub_download('hysts/TADNE-sample-images',
                           f'{size}/{dirname}.tar',
                           repo_type='dataset',
                           use_auth_token=TOKEN)
    return path


def load_deepdanbooru_tag_dict() -> dict[str, int]:
    path = hf_hub_download('hysts/DeepDanbooru',
                           'tags.txt',
                           use_auth_token=TOKEN)
    with open(path) as f:
        tags = [line.strip() for line in f.readlines()]
    return {tag: i for i, tag in enumerate(tags)}


def load_deepdanbooru_predictions(dirname: str) -> np.ndarray:
    path = hf_hub_download('hysts/TADNE-sample-images',
                           f'prediction_results/deepdanbooru/{dirname}.npy',
                           repo_type='dataset',
                           use_auth_token=TOKEN)
    return np.load(path)


def run(
    general_tags: list[str],
    hair_color_tags: list[str],
    hair_style_tags: list[str],
    image_color_tags: list[str],
    score_threshold: float,
    start_index: int,
    nrows: int,
    ncols: int,
    image_size: int,
    min_seed: int,
    max_seed: int,
    dirname: str,
    tarball_path: pathlib.Path,
    deepdanbooru_tag_dict: dict[str, int],
    deepdanbooru_predictions: np.ndarray,
) -> np.ndarray:
    hair_color_tags = [f'{color}_hair' for color in hair_color_tags]

    tags = general_tags + hair_color_tags + hair_style_tags + image_color_tags
    tag_indices = [deepdanbooru_tag_dict[tag] for tag in tags]

    conditions = deepdanbooru_predictions[:, tag_indices] > score_threshold
    image_indices = np.arange(len(deepdanbooru_predictions))
    image_indices = image_indices[conditions.all(axis=1)]

    start_index = int(start_index)
    num = nrows * ncols
    seeds = []
    images = []
    dummy = np.ones((image_size, image_size, 3), dtype=np.uint8) * 255
    with tarfile.TarFile(tarball_path) as tar_file:
        for index in range(start_index, start_index + num):
            if index >= len(image_indices):
                seeds.append(-1)
                images.append(dummy)
                continue
            image_index = image_indices[index]
            seeds.append(image_index)
            member = tar_file.getmember(f'{dirname}/{image_index:07d}.jpg')
            with tar_file.extractfile(member) as f:
                data = io.BytesIO(f.read())
            image = PIL.Image.open(data)
            image = np.asarray(image)
            images.append(image)
    res = np.asarray(images).reshape(nrows, ncols, image_size, image_size,
                                     3).transpose(0, 2, 1, 3, 4).reshape(
                                         nrows * image_size,
                                         ncols * image_size, 3)
    seeds = np.asarray(seeds).reshape(nrows, ncols)

    return len(image_indices), res, seeds


def main():
    gr.close_all()

    args = parse_args()

    image_size = 128
    min_seed = 0
    max_seed = 99999
    dirname = '0-99999'
    tarball_path = download_image_tarball(image_size, dirname)

    deepdanbooru_tag_dict = load_deepdanbooru_tag_dict()
    deepdanbooru_predictions = load_deepdanbooru_predictions(dirname)

    func = functools.partial(
        run,
        image_size=image_size,
        min_seed=min_seed,
        max_seed=max_seed,
        dirname=dirname,
        tarball_path=tarball_path,
        deepdanbooru_tag_dict=deepdanbooru_tag_dict,
        deepdanbooru_predictions=deepdanbooru_predictions,
    )
    func = functools.update_wrapper(func, run)

    gr.Interface(
        func,
        [
            gr.inputs.CheckboxGroup([
                '1girl',
                '1boy',
                'multiple_girls',
                'multiple_boys',
            ],
                                    label='General'),
            gr.inputs.CheckboxGroup([
                'aqua',
                'black',
                'blonde',
                'blue',
                'brown',
                'green',
                'grey',
                'orange',
                'pink',
                'purple',
                'red',
                'silver',
                'white',
            ],
                                    label='Hair Color'),
            gr.inputs.CheckboxGroup([
                'bangs',
                'curly_hair',
                'long_hair',
                'medium_hair',
                'messy_hair',
                'short_hair',
                'straight_hair',
                'twintails',
            ],
                                    label='Hair Style'),
            gr.inputs.CheckboxGroup([
                'greyscale',
                'monochrome',
            ],
                                    label='Image Color'),
            gr.inputs.Slider(0,
                             1,
                             step=0.1,
                             default=0.5,
                             label='DeepDanbooru Score Threshold'),
            gr.inputs.Number(default=0, label='Start Index'),
            gr.inputs.Slider(1, 10, step=1, default=2, label='Number of Rows'),
            gr.inputs.Slider(
                1, 10, step=1, default=5, label='Number of Columns'),
        ],
        [
            gr.outputs.Textbox(type='number', label='Number of Found Images'),
            gr.outputs.Image(type='numpy', label='Output'),
            gr.outputs.Dataframe(type='numpy', label='Seed'),
        ],
        title=TITLE,
        description=DESCRIPTION,
        article=ARTICLE,
        theme=args.theme,
        allow_screenshot=args.allow_screenshot,
        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()