app.py

#!/usr/bin/env python

from __future__ import annotations

import functools
import json
import os
import pathlib
import tarfile
from typing import Callable

import gradio as gr
import huggingface_hub
import PIL.Image
import torch
import torchvision.transforms as T

DESCRIPTION = "# [RF5/danbooru-pretrained](https://github.com/RF5/danbooru-pretrained)"

MODEL_REPO = "public-data/danbooru-pretrained"


def load_sample_image_paths() -> list[pathlib.Path]:
    image_dir = pathlib.Path("images")
    if not image_dir.exists():
        dataset_repo = "hysts/sample-images-TADNE"
        path = huggingface_hub.hf_hub_download(dataset_repo, "images.tar.gz", repo_type="dataset")
        with tarfile.open(path) as f:
            f.extractall()
    return sorted(image_dir.glob("*"))


def load_model(device: torch.device) -> torch.nn.Module:
    path = huggingface_hub.hf_hub_download(MODEL_REPO, "resnet50-13306192.pth")
    state_dict = torch.load(path)
    model = torch.hub.load("RF5/danbooru-pretrained", "resnet50", pretrained=False)
    model.load_state_dict(state_dict)
    model.to(device)
    model.eval()
    return model


def load_labels() -> list[str]:
    path = huggingface_hub.hf_hub_download(MODEL_REPO, "class_names_6000.json")
    with open(path) as f:
        labels = json.load(f)
    return labels


@torch.inference_mode()
def predict(
    image: PIL.Image.Image,
    score_threshold: float,
    transform: Callable,
    device: torch.device,
    model: torch.nn.Module,
    labels: list[str],
) -> dict[str, float]:
    data = transform(image)
    data = data.to(device).unsqueeze(0)
    preds = model(data)[0]
    preds = torch.sigmoid(preds)
    preds = preds.cpu().numpy().astype(float)

    res = dict()
    for prob, label in zip(preds.tolist(), labels):
        if prob < score_threshold:
            continue
        res[label] = prob
    return res


image_paths = load_sample_image_paths()
examples = [[path.as_posix(), 0.4] for path in image_paths]

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = load_model(device)
labels = load_labels()

transform = T.Compose(
    [
        T.Resize(360),
        T.ToTensor(),
        T.Normalize(mean=[0.7137, 0.6628, 0.6519], std=[0.2970, 0.3017, 0.2979]),
    ]
)

fn = functools.partial(predict, transform=transform, device=device, model=model, labels=labels)

with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    with gr.Row():
        with gr.Column():
            image = gr.Image(label="Input", type="pil")
            threshold = gr.Slider(label="Score Threshold", minimum=0, maximum=1, step=0.05, value=0.4)
            run_button = gr.Button()
        with gr.Column():
            result = gr.Label(label="Output")

    inputs = [image, threshold]
    gr.Examples(
        examples=examples,
        inputs=inputs,
        outputs=result,
        fn=fn,
        cache_examples=os.getenv("CACHE_EXAMPLES") == "1",
    )
    run_button.click(
        fn=fn,
        inputs=inputs,
        outputs=result,
        api_name="predict",
    )

if __name__ == "__main__":
    demo.queue(max_size=15).launch()