app.py

from transformers import pipeline
from imgutils.data import rgb_encode, load_image
from onnx_ import _open_onnx_model
from PIL import Image
import gradio as gr
import numpy as np
import os
import requests
import torch
import json

def _img_encode(image, size=(384,384), normalize=(0.5,0.5)):
    image = image.resize(size, Image.BILINEAR)
    data = rgb_encode(image, order_='CHW')

    if normalize is not None:
        mean_, std_ = normalize
        mean = np.asarray([mean_]).reshape((-1, 1, 1))
        std = np.asarray([std_]).reshape((-1, 1, 1))
        data = (data - mean) / std

    return data.astype(np.float32)
    
nsfw_tf = pipeline(model="carbon225/vit-base-patch16-224-hentai")

if not os.path.exists("timm.onnx"):
    open("timm.onnx", "wb").write(
        requests.get(
            "https://huggingface.co/deepghs/anime_rating/resolve/main/caformer_s36_plus/model.onnx"
        ).content
    )
    open("timmcfg.json", "wb").write(
        requests.get(
            "https://huggingface.co/deepghs/anime_rating/resolve/main/caformer_s36_plus/meta.json"
        ).content
    )
else:
    print("Model already exists, skipping redownload")

with open("timmcfg.json") as file:
    tm_cfg = json.load(file)

nsfw_tm = _open_onnx_model("timm.onnx")

def launch(img):
    weight = 0
    img = img.convert('RGB')
    tm_image = load_image(img, mode='RGB')
    tm_input_ = _img_encode(tm_image, size=(256, 256))[None, ...]
    tm_items, = nsfw_tm.run(['output'], {'input': tm_input_})
    tm_output = sorted(list(zip(tm_cfg["labels"], map(lambda x: x.item(), tm_items[0]))), key=lambda x: x[1], reverse=True)[0][0]
    
    match tm_output:
        case "safe":
            weight -= 1
        case "r15":
            weight += 2
        case "r18":
            weight += 2

    tf_output = nsfw_tf(img)[0]["label"]

    match tf_output:
        case "safe":
            weight -= 1
        case "suggestive":
            weight += 1
        case "r18":
            weight += 2

    print(sorted(list(zip(tm_cfg["labels"], map(lambda x: x.item(), tm_items[0]))), key=lambda x: x[1], reverse=True), tf_output)
    return weight > 0

app = gr.Interface(fn=launch, inputs="pil", outputs="text")
app.launch()