app.py

import os
import numpy as np
import codecs
import torch
import torchvision.transforms as transforms
import gradio as gr

from PIL import Image

from unetplusplus import NestedUNet

torch.manual_seed(0)

if torch.cuda.is_available():
    torch.backends.cudnn.deterministic = True

# Device
DEVICE = "cpu"
print(DEVICE)

# Load color map
cmap = np.load("cmap.npy")

# Make directories
os.system("mkdir ./models")

# Get model weights
if not os.path.exists("./models/masksupnyu39.31d.pth"):
    os.system(
        "wget -O ./models/masksupnyu39.31d.pth https://github.com/hasibzunair/masksup-segmentation/releases/download/v0.1/masksupnyu39.31iou.pth"
    )

# Load model
model = NestedUNet(num_classes=40)
checkpoint = torch.load(
    "./models/masksupnyu39.31d.pth", map_location=torch.device("cpu")
)
model.load_state_dict(checkpoint)
model = model.to(DEVICE)
model.eval()


# Main inference function
def inference(img_path):
    image = Image.open(img_path).convert("RGB")
    transforms_image = transforms.Compose(
        [
            transforms.Resize((224, 224)),
            transforms.CenterCrop((224, 224)),
            transforms.ToTensor(),
            transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
        ]
    )

    image = transforms_image(image)
    image = image[None, :]  # batch dimension

    # Predict
    with torch.no_grad():
        output = torch.sigmoid(model(image.to(DEVICE).float()))
        output = (
            torch.softmax(output, dim=1)
            .argmax(dim=1)[0]
            .float()
            .cpu()
            .numpy()
            .astype(np.uint8)
        )
        pred = cmap[output]
    return pred


# App
title = "Masked Supervised Learning for Semantic Segmentation"
description = codecs.open("description.html", "r", "utf-8").read()
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2210.00923' target='_blank'>Masked Supervised Learning for Semantic Segmentation</a> | <a href='https://github.com/hasibzunair/masksup-segmentation' target='_blank'>Github</a></p>"

gr.Interface(
    inference,
    gr.inputs.Image(type="filepath", label="Input Image"),
    gr.outputs.Image(type="numpy", label="Predicted Output"),
    examples=[
        "./sample_images/a.png",
        "./sample_images/b.png",
        "./sample_images/c.png",
        "./sample_images/d.png",
    ],
    title=title,
    description=description,
    article=article,
    allow_flagging=False,
    analytics_enabled=False,
).launch(debug=True, enable_queue=True)