PIFu/spaces.py

import os
from xml.etree.ElementPath import ops
try:
    os.system("pip install --upgrade  torch==1.11.0+cu113 torchvision==0.12.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html")
except Exception as e:
    print(e)

from pydoc import describe
from huggingface_hub import hf_hub_download
import gradio as gr
import os
from datetime import datetime
from PIL import Image
import torch
import torchvision
import skimage
import paddlehub
import numpy as np
from lib.options import BaseOptions
from apps.crop_img import process_img
from apps.eval import Evaluator
from types import SimpleNamespace
import trimesh
import glob

os.environ["CUDA_VISIBLE_DEVICES"]="0"

print(
    "torch: ", torch.__version__,
    "\ntorchvision: ", torchvision.__version__,
    "\nskimage:", skimage.__version__
)

net_C = hf_hub_download("radames/PIFu-upright-standing", filename="net_C")
net_G = hf_hub_download("radames/PIFu-upright-standing", filename="net_G")


opt = BaseOptions()
opts = opt.parse_to_dict()
opts['batch_size'] = 1
opts['mlp_dim'] = [257, 1024, 512, 256, 128, 1]
opts['mlp_dim_color'] = [513, 1024, 512, 256, 128, 3]
opts['num_stack'] = 4
opts['num_hourglass'] = 2
opts['resolution'] = 128
opts['hg_down'] = 'ave_pool'
opts['norm'] = 'group'
opts['norm_color'] = 'group'
opts['load_netG_checkpoint_path'] = net_G
opts['load_netC_checkpoint_path'] = net_C
opts['results_path'] = "./results"
opts['name'] = "spaces_demo"
opts = SimpleNamespace(**opts)
evaluator = Evaluator(opts)
bg_remover_model = paddlehub.Module(name="U2Net")


def process(img_path):
    base = os.path.basename(img_path)
    img_name = os.path.splitext(base)[0]
    print("\n\n\nStarting Process", datetime.now())
    print("image name", img_name)
    img_raw = Image.open(img_path).convert('RGB')

    img = img_raw.resize(
        (800, int(800 * img_raw.size[1] / img_raw.size[0])),
        Image.Resampling.LANCZOS)

    try:
        # remove background
        print("Removing Background")
        masks = bg_remover_model.Segmentation(
            images=[np.array(img)],
            paths=None,
            batch_size=1,
            input_size=320,
            output_dir='./PIFu/inputs',
            visualization=False)
        mask = masks[0]["mask"]
        front = masks[0]["front"]
    except Exception as e:
        print(e)

    print("Aliging mask with input training image")
    print("Not aligned", front.shape, mask.shape)
    img_new, msk_new = process_img(front, mask)
    print("Aligned", img_new.shape, msk_new.shape)

    try:
        time = datetime.now()
        data = evaluator.load_image_from_memory(img_new, msk_new, img_name)
        print("Evaluating via PIFu", time)
        evaluator.eval(data, True)
        print("Success Evaluating via PIFu", datetime.now() - time)
        result_path = f'./{opts.results_path}/{opts.name}/result_{img_name}'
    except Exception as e:
        print("Error evaluating via PIFu", e)

    try:
        mesh = trimesh.load(result_path + '.obj')
        # flip mesh
        mesh.apply_transform([[-1, 0, 0, 0],
                              [0, 1, 0, 0],
                              [0, 0, -1, 0],
                              [0, 0, 0, 1]])
        mesh.export(file_obj=result_path + '.glb')
        result_gltf = result_path + '.glb'
        return [result_gltf, result_gltf]

    except Exception as e:
        print("error generating MESH", e)


examples = sorted(glob.glob('examples/*.png'))
description = '''
# PIFu Clothed Human Digitization
### PIFu: Pixel-Aligned Implicit Function for High-Resolution Clothed Human Digitization
<base target="_blank">

This is a demo for <a href="https://github.com/shunsukesaito/PIFu" target="_blank"> PIFu model </a>.
The pre-trained model has the following warning:
> Warning: The released model is trained with mostly upright standing scans with weak perspectie projection and the pitch angle of 0 degree. Reconstruction quality may degrade for images highly deviated from trainining data.

**The inference takes about 180seconds for a new image.**

<details>
<summary>More</summary>

#### Image Credits

* Julien and Clem
* [StyleGAN Humans](https://huggingface.co/spaces/hysts/StyleGAN-Human)
* [Renderpeople: Dennis](https://renderpeople.com)


#### More
* https://phorhum.github.io/
* https://github.com/yuliangxiu/icon
* https://shunsukesaito.github.io/PIFuHD/

</details>
'''

iface = gr.Interface(
    fn=process,
    description=description,
    inputs=gr.inputs.Image(type="filepath", label="Input Image"),
    outputs=[
        gr.outputs.Image3D(
            clear_color=[0.0, 0.0, 0.0, 0.0],  label="3D Model"),
        gr.outputs.File(label="Download 3D Model")
    ],
    examples=examples,
    allow_flagging="never",
)

if __name__ == "__main__":
    iface.launch(debug=True, enable_queue=False, cache_examples=True)