app.py

from __future__ import annotations

import argparse
import os
import pathlib
import subprocess

if os.getenv('SYSTEM') == 'spaces':
    import mim

    mim.uninstall('mmcv-full', confirm_yes=True)
    mim.install('mmcv-full==1.4.3', is_yes=True)

    subprocess.call('pip uninstall -y opencv-python'.split())
    subprocess.call('pip uninstall -y opencv-python-headless'.split())
    subprocess.call('pip install opencv-python-headless==4.5.5.64'.split())
    subprocess.call('pip install pycocotools'.split())
    subprocess.call("pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu102/torch1.9/index.html".split())
#    subprocess.call("pip install git+https://github.com/c-liangyu/OpenPSG.git@dev_apis".split())
    subprocess.call("pip install git+https://github.com/Jingkang50/OpenPSG.git@hugging_face_demo".split())
    subprocess.call("pip install git+https://github.com/cocodataset/panopticapi.git".split())

import cv2
import gradio as gr
import numpy as np

from mmdet.apis import init_detector, inference_detector
from utils import make_gif, show_result
from mmcv import Config
import openpsg

DESCRIPTION = '''# ECCV'22 | Panoptic Scene Graph Generation


🚀 🚀 🚀 This is an official demo for our ECCV'22 paper: [Panoptic Scene Graph Generation](https://psgdataset.org/). Please star our [codebase](https://github.com/Jingkang50/OpenPSG) if you find it useful / interesting.

📢 📢 📢 **News:** The PSG Challenge (prize pool 🤑 **US$150K** 🤑) is now available on [International Algorithm Case Competition](https://www.cvmart.net/race/10349/base?organic_url=https%3A%2F%2Fhf.space%2F) and [ECCV'22 SenseHuman Workshop](https://sense-human.github.io/)!

🔍 🔍 🔍 Check out the [news section](https://github.com/Jingkang50/OpenPSG#updates) in our [GitHub repo](https://github.com/Jingkang50/OpenPSG) for more details. Everyone around the world is welcome to participant and explore the comprehensive scene understanding!

🎯 🎯 🎯 The PSG Development Team is currently focusing on **(1) 🧙‍♂️ Next-Generation PSG Models**, **(2) 🕵️‍♀️ Relation-Aware Visual Reasoning from PSG Models**, and **(3) 🎨 Relation-Aware Image Generation from Scene Graph and Caption**. If you are also interested in the related researches, please reach out and contact us!

Inference takes 10-30 seconds per image. The model is PSGTR (60 epochs). You can upload your own pictures or select the examples below to play.
The demo will output a GIF to show the first 10 "subject-verb-object" relations, with the subject and object being grounded by segmentation masks. 
A gallery is attached below for reference.

'''
FOOTER = '<img id="visitor-badge" src="https://visitor-badge.glitch.me/badge?page_id=c-liangyu.openpsg" alt="visitor badge" />'

def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser()
    parser.add_argument('--device', type=str, default='cpu')
    parser.add_argument('--theme', type=str)
    parser.add_argument('--share', action='store_true')
    parser.add_argument('--port', type=int)
    parser.add_argument('--disable-queue',
                        dest='enable_queue',
                        action='store_false')
    return parser.parse_args()


def update_input_image(image: np.ndarray) -> dict:
    if image is None:
        return gr.Image.update(value=None)
    scale = 800 / max(image.shape[:2])
    if scale < 1:
        image = cv2.resize(image, None, fx=scale, fy=scale)
    return gr.Image.update(value=image)


def set_example_image(example: list) -> dict:
    return gr.Image.update(value=example[0])

class Model:
    def __init__(self, model_name, device='cpu'):
        model_ckt ='OpenPSG/checkpoints/epoch_60.pth'
        cfg = Config.fromfile('OpenPSG/configs/psgtr/psgtr_r50_psg_inference.py')
        self.model = init_detector(cfg, model_ckt, device=device)

    def infer(self, input_image, num_rel):
        result = inference_detector(self.model, input_image)
        displays = show_result(input_image,
                            result,
                            is_one_stage=True,
                            num_rel=num_rel,
                            show=True
                            )
        gif = make_gif(displays[:10] if len(displays) > 10 else displays)
        return gif, displays


def main():
    args = parse_args()

    with gr.Blocks(theme=args.theme, css='style.css') as demo:
        
        model = Model('psgtr', device=args.device)
        
        gr.Markdown(DESCRIPTION)

        with gr.Row():
            with gr.Column():
                with gr.Row():
                    input_image = gr.Image(label='Input Image', type='numpy')
                with gr.Group():
                    with gr.Row():
                        num_rel = gr.Slider(
                            5,
                            100,
                            step=5,
                            value=20,
                            label='Number of Relations')
                with gr.Row():
                    run_button = gr.Button(value='Run')
            with gr.Column():
                with gr.Row():
                    gif = gr.Image(label='Top Relations')
                with gr.Row():
                    displays = gr.Gallery(label='PSGTR Result', type='numpy')

        with gr.Row():
            paths = sorted(pathlib.Path('images').rglob('*.jpg'))
            example_images = gr.Dataset(components=[input_image],
                                        samples=[[path.as_posix()]
                                                 for path in paths])

        gr.Markdown(FOOTER)

        input_image.change(fn=update_input_image,
                           inputs=input_image,
                           outputs=input_image)
        
        run_button.click(fn=model.infer,
                         inputs=[
                            input_image, num_rel
                         ],
                         outputs=[gif, displays])

        example_images.click(fn=set_example_image,
                             inputs=example_images,
                             outputs=input_image)

    demo.launch(
        enable_queue=args.enable_queue,
        server_port=args.port,
        share=args.share,
    )


if __name__ == '__main__':
    main()