app.py

#!/usr/bin/env python

from __future__ import annotations

import os
import datetime
import pathlib
import shlex
import subprocess
import sys
from typing import Generator, Optional

import trimesh
import spaces
import gradio as gr

# from model import Model

sys.path.append('TEXTurePaper')

from src.configs.train_config import GuideConfig, LogConfig, TrainConfig
from src.training.trainer import TEXTure


class Model:
    def __init__(self):
        self.max_num_faces = 100000

    def load_config(self, shape_path: str, text: str, seed: int,
                    guidance_scale: float) -> TrainConfig:
        text += ', {} view'

        log = LogConfig(exp_name=self.gen_exp_name())
        guide = GuideConfig(text=text)
        guide.background_img = 'TEXTurePaper/textures/brick_wall.png'
        guide.shape_path = 'TEXTurePaper/shapes/spot_triangulated.obj'
        config = TrainConfig(log=log, guide=guide)

        config.guide.shape_path = shape_path
        config.optim.seed = seed
        config.guide.guidance_scale = guidance_scale
        return config

    def gen_exp_name(self) -> str:
        now = datetime.datetime.now()
        return now.strftime('%Y-%m-%d-%H-%M-%S')

    def check_num_faces(self, path: str) -> bool:
        with open(path) as f:
            lines = [line for line in f.readlines() if line.startswith('f')]
        return len(lines) <= self.max_num_faces

    def zip_results(self, exp_dir: pathlib.Path) -> str:
        mesh_dir = exp_dir / 'mesh'
        out_path = f'{exp_dir.name}.zip'
        subprocess.run(shlex.split(f'zip -r {out_path} {mesh_dir}'))
        return out_path

    def run(
        self, shape_path: str, text: str, seed: int, guidance_scale: float
    ) -> Generator[tuple[list[str], Optional[str], Optional[str], str], None,
                   None]:
        if not shape_path.endswith('.obj'):
            raise gr.Error('The input file is not .obj file.')
        if not self.check_num_faces(shape_path):
            raise gr.Error('The number of faces is over 100,000.')

        config = self.load_config(shape_path, text, seed, guidance_scale)
        trainer = TEXTure(config)

        trainer.mesh_model.train()

        total_steps = len(trainer.dataloaders['train'])
        for step, data in enumerate(trainer.dataloaders['train'], start=1):
            trainer.paint_step += 1
            trainer.paint_viewpoint(data)
            trainer.evaluate(trainer.dataloaders['val'],
                             trainer.eval_renders_path)
            trainer.mesh_model.train()

            sample_image_dir = config.log.exp_dir / 'vis' / 'eval'
            sample_image_paths = sorted(
                sample_image_dir.glob(f'step_{trainer.paint_step:05d}_*.jpg'))
            sample_image_paths = [
                path.as_posix() for path in sample_image_paths
            ]
            yield sample_image_paths, None, None, f'{step}/{total_steps}'

        trainer.mesh_model.change_default_to_median()

        save_dir = trainer.exp_path / 'mesh'
        save_dir.mkdir(exist_ok=True, parents=True)
        trainer.mesh_model.export_mesh(save_dir)
        model_path = save_dir / 'mesh.obj'
        mesh = trimesh.load(model_path)
        mesh_path = save_dir / 'mesh.glb'
        mesh.export(mesh_path, file_type='glb')

        zip_path = self.zip_results(config.log.exp_dir)
        yield sample_image_paths, mesh_path.as_posix(), zip_path, 'Done!'


@spaces.GPU
def main():
    DESCRIPTION = '''# [TEXTure](https://github.com/TEXTurePaper/TEXTurePaper)
    
    - This demo only accepts as input `.obj` files with less than 100,000 faces.
    - Inference takes about 10 minutes on a T4 GPU.
    '''
    if (SPACE_ID := os.getenv('SPACE_ID')) is not None:
        DESCRIPTION += f'\n<p>For faster inference without waiting in queue, you may duplicate the space and upgrade to GPU in settings. <a href="https://huggingface.co/spaces/{SPACE_ID}?duplicate=true"><img style="display: inline; margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space" /></a></p>'

    model = Model()

    with gr.Blocks(css='style.css') as demo:
        gr.Markdown(DESCRIPTION)
        with gr.Row():
            with gr.Column():
                input_shape = gr.Model3D(label='Input 3D mesh')
                text = gr.Text(label='Text')
                seed = gr.Slider(label='Seed',
                                 minimum=0,
                                 maximum=100000,
                                 value=3,
                                 step=1)
                guidance_scale = gr.Slider(label='Guidance scale',
                                           minimum=0,
                                           maximum=50,
                                           value=7.5,
                                           step=0.1)
                run_button = gr.Button('Run')
            with gr.Column():
                progress_text = gr.Text(label='Progress')
                with gr.Tabs():
                    with gr.TabItem(label='Images from each viewpoint'):
                        viewpoint_images = gr.Gallery(show_label=False, columns=4)
                    with gr.TabItem(label='Result 3D model'):
                        result_3d_model = gr.Model3D(show_label=False)
                    with gr.TabItem(label='Output mesh file'):
                        output_file = gr.File(show_label=False)
        with gr.Row():
            examples = [
                ['shapes/dragon1.obj', 'a photo of a dragon', 0, 7.5],
                ['shapes/dragon2.obj', 'a photo of a dragon', 0, 7.5],
                ['shapes/eagle.obj', 'a photo of an eagle', 0, 7.5],
                ['shapes/napoleon.obj', 'a photo of Napoleon Bonaparte', 3, 7.5],
                ['shapes/nascar.obj', 'A next gen nascar', 2, 10],
            ]
            gr.Examples(examples=examples,
                        inputs=[
                            input_shape,
                            text,
                            seed,
                            guidance_scale,
                        ],
                        outputs=[
                            result_3d_model,
                            output_file,
                        ],
                        cache_examples=False)

        run_button.click(fn=model.run,
                         inputs=[
                             input_shape,
                             text,
                             seed,
                             guidance_scale,
                         ],
                         outputs=[
                             viewpoint_images,
                             result_3d_model,
                             output_file,
                             progress_text,
                         ])

    demo.queue(max_size=5).launch(debug=True)


if __name__ == '__main__':
    main()