app.py

#!/usr/bin/env python

from __future__ import annotations

import functools
import sys

import gradio as gr
import huggingface_hub
import PIL.Image
import spaces
import torch
import torch.nn as nn

sys.path.insert(0, "Anime2Sketch")

from data import read_img_path, tensor_to_img
from model import UnetGenerator

TITLE = "Anime2Sketch"
DESCRIPTION = "https://github.com/Mukosame/Anime2Sketch"


def load_model(device: torch.device) -> nn.Module:
    norm_layer = functools.partial(nn.InstanceNorm2d, affine=False, track_running_stats=False)
    model = UnetGenerator(3, 1, 8, 64, norm_layer=norm_layer, use_dropout=False)

    path = huggingface_hub.hf_hub_download("public-data/Anime2Sketch", "netG.pth")
    ckpt = torch.load(path)
    for key in list(ckpt.keys()):
        if "module." in key:
            ckpt[key.replace("module.", "")] = ckpt[key]
            del ckpt[key]
    model.load_state_dict(ckpt)
    model.to(device)
    model.eval()
    return model


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


@spaces.GPU
@torch.inference_mode()
def run(image_file: str, load_size: int = 512) -> PIL.Image.Image:
    tensor, orig_size = read_img_path(image_file, load_size)
    tensor = tensor.to(device)
    out = model(tensor)
    res = tensor_to_img(out)
    res = PIL.Image.fromarray(res)
    res = res.resize(orig_size, PIL.Image.Resampling.BICUBIC)
    return res


demo = gr.Interface(
    fn=run,
    inputs=gr.Image(label="Input", type="filepath"),
    outputs=gr.Image(label="Output"),
    examples=["Anime2Sketch/test_samples/madoka.jpg"],
    title=TITLE,
    description=DESCRIPTION,
)

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