app.py

import gradio as gr
from PIL import Image
import numpy as np
from gradio import components
import torchvision
from torchvision.models.detection import (
    maskrcnn_resnet50_fpn,
    MaskRCNN_ResNet50_FPN_Weights,
)
import torchvision.transforms.functional as F
import torch
from torchvision.utils import draw_segmentation_masks

weights = MaskRCNN_ResNet50_FPN_Weights.DEFAULT
transforms = weights.transforms()

model = maskrcnn_resnet50_fpn(weights=weights, progress=False)
model = model.eval()


def segment_and_show(image):
    # abc
    input_image = Image.fromarray(image)
    input_tensor = torch.tensor(np.array(input_image))
    input_tensor = input_tensor.permute(2, 0, 1)
    input_image = transforms(input_image)
    output = model([input_image])[0]
    proba_threshold = 0.5
    masks = output["masks"] > proba_threshold
    masks = masks.squeeze(1)
    image_with_segmasks = draw_segmentation_masks(input_tensor, masks, alpha=0.7)
    return np.array(F.to_pil_image(image_with_segmasks))


default_image = Image.open("demo.jpeg")

iface = gr.Interface(
    fn=segment_and_show,
    inputs=components.Image(value=default_image, sources=["upload", "clipboard"]),
    outputs=components.Image(type="pil"),
    title="Urban Autonomy Instance Segmentation Demo",
    description="Upload an image or use the default to see the instance segmentation model in action.",
)

if __name__ == "__main__":
    iface.launch()