app.py

#!/usr/bin/env python

import os
import pathlib
import shlex
import subprocess
import tarfile

if os.getenv("SYSTEM") == "spaces":
    subprocess.run(shlex.split("pip install git+https://github.com/facebookresearch/detectron2@v0.6"), check=True)  # noqa: S603
    subprocess.run(shlex.split("pip install git+https://github.com/aim-uofa/AdelaiDet@7bf9d87"), check=True)  # noqa: S603
    subprocess.run(shlex.split("pip install Pillow==9.5.0"), check=True)  # noqa: S603

import gradio as gr
import huggingface_hub
import numpy as np
import torch
from adet.config import get_cfg
from detectron2.data.detection_utils import read_image
from detectron2.engine.defaults import DefaultPredictor
from detectron2.utils.visualizer import Visualizer

DESCRIPTION = "# [Yet-Another-Anime-Segmenter](https://github.com/zymk9/Yet-Another-Anime-Segmenter)"

MODEL_REPO = "public-data/Yet-Another-Anime-Segmenter"


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()  # noqa: S202
    return sorted(image_dir.glob("*"))


def load_model(device: torch.device) -> DefaultPredictor:
    config_path = huggingface_hub.hf_hub_download(MODEL_REPO, "SOLOv2.yaml")
    model_path = huggingface_hub.hf_hub_download(MODEL_REPO, "SOLOv2.pth")
    cfg = get_cfg()
    cfg.merge_from_file(config_path)
    cfg.MODEL.WEIGHTS = model_path
    cfg.MODEL.DEVICE = device.type
    cfg.freeze()
    return DefaultPredictor(cfg)


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


def predict(
    image_path: str, class_score_threshold: float, mask_score_threshold: float
) -> tuple[np.ndarray, np.ndarray]:
    model.score_threshold = class_score_threshold
    model.mask_threshold = mask_score_threshold
    image = read_image(image_path, format="BGR")
    preds = model(image)
    instances = preds["instances"].to("cpu")

    visualizer = Visualizer(image[:, :, ::-1])
    vis = visualizer.draw_instance_predictions(predictions=instances)
    vis = vis.get_image()

    masked = image.copy()[:, :, ::-1]
    mask = instances.pred_masks.cpu().numpy().astype(int).max(axis=0)
    masked[mask == 0] = 255

    return vis, masked


image_paths = load_sample_image_paths()
examples = [[path, 0.1, 0.5] for path in image_paths]


with gr.Blocks(css_paths="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    with gr.Row():
        with gr.Column():
            image = gr.Image(label="Input", type="filepath")
            class_score_threshold = gr.Slider(label="Score Threshold", minimum=0, maximum=1, step=0.05, value=0.1)
            mask_score_threshold = gr.Slider(label="Mask Score Threshold", minimum=0, maximum=1, step=0.05, value=0.5)
            run_button = gr.Button()
        with gr.Column():
            result_instances = gr.Image(label="Instances")
            result_masked = gr.Image(label="Masked")

    inputs = [image, class_score_threshold, mask_score_threshold]
    outputs = [result_instances, result_masked]
    gr.Examples(
        examples=examples,
        inputs=inputs,
        outputs=outputs,
        fn=predict,
        cache_examples=os.getenv("CACHE_EXAMPLES") == "1",
    )
    run_button.click(
        fn=predict,
        inputs=inputs,
        outputs=outputs,
    )

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