app-onnx.py

#######################################################################################
#
# MIT License
#
# Copyright (c) [2025] [leonelhs@gmail.com]
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
#
#######################################################################################

# This file implements an API endpoint for DIS background image removal system.
# [Self space] - [https://huggingface.co/spaces/leonelhs/removebg]
#
# Source code is based on or inspired by several projects.
# For more details and proper attribution, please refer to the following resources:
#
# - [DIS] - [https://github.com/xuebinqin/DIS]
# - [removebg] - [https://huggingface.co/spaces/gaviego/removebg]
# https://github.com/gaurav0651/dis-bg-remover

from itertools import islice

import cv2
import gradio as gr
import numpy as np
import onnxruntime as ort
from PIL import Image
from huggingface_hub import hf_hub_download

REPO_ID = "leonelhs/removators"

# Load the ONNX model
model_path = hf_hub_download(repo_id=REPO_ID, filename='isnet.onnx')

session = ort.InferenceSession(model_path)

def normalize(image, mean, std):
    """Normalize a numpy image with mean and standard deviation."""
    return (image / 255.0 - mean) / std

def predict(image_path):
    input_size = (1024, 1024)

    img = cv2.imread(image_path, cv2.IMREAD_COLOR)
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)  # Convert from BGR to RGB if using OpenCV

    # If image is grayscale, convert to RGB
    if len(img.shape) == 2:
        img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)

    # Normalize the image using NumPy
    img = img.astype(np.float32)  # Convert to float
    im_normalized = normalize(img, mean=[0.5, 0.5, 0.5], std=[1.0, 1.0, 1.0])

    # Resize the image
    img_resized = cv2.resize(im_normalized, input_size, interpolation=cv2.INTER_LINEAR)
    img_resized = np.transpose(img_resized, (2, 0, 1))  # CHW format
    img_resized = np.expand_dims(img_resized, axis=0)  # Add batch dimension

    # Run inference
    img_resized = img_resized.astype(np.float32)
    ort_inputs = {session.get_inputs()[0].name: img_resized}
    prediction = session.run(None, ort_inputs)

    # Process the model output
    result = prediction[0][0]  # Assuming single output and single batch
    result = np.clip(result, 0, 1)  # Assuming you want to clip the result to [0, 1]
    result = (result * 255).astype(np.uint8)  # Rescale to [0, 255]
    result = np.transpose(result, (1, 2, 0))  # HWC format
    # Resize to original shape
    original_shape = img.shape[:2]
    return cv2.resize(result, (original_shape[1], original_shape[0]), interpolation=cv2.INTER_LINEAR)


def cuts(image):
    mask = predict(image)
    mask = Image.fromarray(mask).convert('L')
    cutted = Image.open(image).convert("RGB")
    cutted.putalpha(mask)
    return [image, cutted], mask

with gr.Blocks(title="DIS") as app:
    navbar = gr.Navbar(visible=True, main_page_name="Workspace")
    gr.Markdown("## Dichotomous Image Segmentation")
    with gr.Row():
        with gr.Column(scale=1):
            inp_image = gr.Image(type="filepath", label="Upload Image")
            btn_predict = gr.Button(variant="primary", value="Remove background")
        with gr.Column(scale=2):
            with gr.Row():
                preview = gr.ImageSlider(type="filepath", label="Comparer")

    btn_predict.click(cuts, inputs=[inp_image], outputs=[preview, inp_image])

with app.route("Readme", "/readme"):
    with open("README.md") as f:
        for line in islice(f, 12, None):
            gr.Markdown(line.strip())


app.launch(share=False, debug=True, show_error=True, mcp_server=True, pwa=True)
app.queue()