# install
import glob
import gradio as gr
import os
import subprocess
curr_dir = os.path.dirname(__file__)
if os.getenv('SYSTEM') == 'spaces':
# subprocess.run('pip install pyembree'.split())
subprocess.run(
'pip install --no-index --no-cache-dir pytorch3d -f https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py38_cu116_pyt1130/download.html'
.split()
)
subprocess.run(
f"cd {curr_dir}/lib/common/libmesh && python setup.py build_ext --inplace".split()
)
subprocess.run(
f"cd {curr_dir}/lib/common/libvoxelize && python setup.py build_ext --inplace".split()
)
subprocess.run(f"cd {curr_dir}".split())
from apps.infer import generate_model, generate_video
# running
description = '''
# Unconstrained & Detailed Clothed Human Digitization (ECON + ControlNet)
### ECON: Explicit Clothed humans Optimized via Normal integration (CVPR 2023, Highlight)
- Colab Notebook
- Blender Plugin
- Docker Image
- Windows Setup
|
|
|
#### Citation
```
@inproceedings{xiu2023econ,
title = {{ECON: Explicit Clothed humans Optimized via Normal integration}},
author = {Xiu, Yuliang and Yang, Jinlong and Cao, Xu and Tzionas, Dimitrios and Black, Michael J.},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2023},
}
```
More
#### Acknowledgments:
- [controlnet-openpose](https://huggingface.co/spaces/diffusers/controlnet-openpose)
- [TEXTure](https://huggingface.co/spaces/TEXTurePaper/TEXTure)
#### Image Credits
* [Pinterest](https://www.pinterest.com/search/pins/?q=parkour&rs=sitelinks_searchbox)
#### Related works
* [ICON @ MPI-IS](https://icon.is.tue.mpg.de/)
* [MonoPort @ USC](https://xiuyuliang.cn/monoport)
* [Phorhum @ Google](https://phorhum.github.io/)
* [PIFuHD @ Meta](https://shunsukesaito.github.io/PIFuHD/)
* [PaMIR @ Tsinghua](http://www.liuyebin.com/pamir/pamir.html)
Generate pose & prompt-guided images / Upload photos / Use examples → Submit Image (~2min) → Generate Video (~2min)
'''
from controlnet_aux import OpenposeDetector
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
from diffusers import UniPCMultistepScheduler
import gradio as gr
import torch
import base64
from io import BytesIO
from PIL import Image
# live conditioning
canvas_html = ""
load_js = """
async () => {
const url = "https://huggingface.co/datasets/radames/gradio-components/raw/main/pose-gradio.js"
fetch(url)
.then(res => res.text())
.then(text => {
const script = document.createElement('script');
script.type = "module"
script.src = URL.createObjectURL(new Blob([text], { type: 'application/javascript' }));
document.head.appendChild(script);
});
}
"""
get_js_image = """
async (image_in_img, prompt, image_file_live_opt, live_conditioning) => {
const canvasEl = document.getElementById("canvas-root");
const data = canvasEl? canvasEl._data : null;
return [image_in_img, prompt, image_file_live_opt, data]
}
"""
# Constants
low_threshold = 100
high_threshold = 200
# Models
pose_model = OpenposeDetector.from_pretrained("lllyasviel/ControlNet")
controlnet = ControlNetModel.from_pretrained(
"lllyasviel/sd-controlnet-openpose", torch_dtype=torch.float16
)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
"runwayml/stable-diffusion-v1-5",
controlnet=controlnet,
safety_checker=None,
torch_dtype=torch.float16
)
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
# This command loads the individual model components on GPU on-demand. So, we don't
# need to explicitly call pipe.to("cuda").
pipe.enable_model_cpu_offload()
# xformers
pipe.enable_xformers_memory_efficient_attention()
# Generator seed,
generator = torch.manual_seed(0)
hint_prompts = '''
Hints:
best quality, extremely detailed, solid color background,
super detail, high detail, edge lighting, soft focus,
light and dark contrast, 8k, high detail, edge lighting,
3d, c4d, blender, oc renderer, ultra high definition, 3d rendering
'''
def get_pose(image):
return pose_model(image)
# def generate_texture(input_shape, text, seed, guidance_scale):
# iface = gr.Interface.load("spaces/TEXTurePaper/TEXTure")
# output_shape = iface(input_shape, text, seed, guidance_scale)
# return output_shape
def generate_images(image, prompt, image_file_live_opt='file', live_conditioning=None):
if image is None and 'image' not in live_conditioning:
raise gr.Error("Please provide an image")
try:
if image_file_live_opt == 'file':
pose = get_pose(image)
elif image_file_live_opt == 'webcam':
base64_img = live_conditioning['image']
image_data = base64.b64decode(base64_img.split(',')[1])
pose = Image.open(BytesIO(image_data)).convert('RGB').resize((512, 512))
output = pipe(
prompt,
pose,
generator=generator,
num_images_per_prompt=3,
num_inference_steps=20,
)
all_outputs = []
all_outputs.append(pose)
for image in output.images:
all_outputs.append(image)
return all_outputs, all_outputs
except Exception as e:
raise gr.Error(str(e))
def toggle(choice):
if choice == "file":
return gr.update(visible=True, value=None), gr.update(visible=False, value=None)
elif choice == "webcam":
return gr.update(visible=False, value=None), gr.update(visible=True, value=canvas_html)
examples_pose = glob.glob('examples/pose/*')
examples_cloth = glob.glob('examples/cloth/*')
default_step = 50
with gr.Blocks() as demo:
gr.Markdown(description)
out_lst = []
with gr.Row():
with gr.Column():
with gr.Row():
live_conditioning = gr.JSON(value={}, visible=False)
with gr.Column():
image_file_live_opt = gr.Radio(["file", "webcam"],
value="file",
label="How would you like to upload your image?")
with gr.Row():
image_in_img = gr.Image(
source="upload", visible=True, type="pil", label="Image for Pose"
)
canvas = gr.HTML(None, elem_id="canvas_html", visible=False)
image_file_live_opt.change(
fn=toggle,
inputs=[image_file_live_opt],
outputs=[image_in_img, canvas],
queue=False
)
prompt = gr.Textbox(
label="Enter your prompt to synthesise the image",
max_lines=10,
placeholder="best quality, extremely detailed",
)
gr.Markdown(hint_prompts)
with gr.Column():
gallery = gr.Gallery().style(grid=[2], height="auto")
gallery_cache = gr.State()
inp = gr.Image(type="filepath", label="Input Image for ECON")
fitting_step = gr.inputs.Slider(
10, 100, step=10, label='Fitting steps', default=default_step
)
with gr.Row():
btn_sample = gr.Button("Generate Image")
btn_submit = gr.Button("Submit Image (~2min)")
btn_sample.click(
fn=generate_images,
inputs=[image_in_img, prompt, image_file_live_opt, live_conditioning],
outputs=[gallery, gallery_cache],
_js=get_js_image
)
def get_select_index(cache, evt: gr.SelectData):
return cache[evt.index]
gallery.select(
fn=get_select_index,
inputs=[gallery_cache],
outputs=[inp],
)
with gr.Row():
gr.Examples(
examples=list(examples_pose),
inputs=[inp],
cache_examples=False,
fn=generate_model,
outputs=out_lst,
label="Hard Pose Exampels"
)
gr.Examples(
examples=list(examples_cloth),
inputs=[inp],
cache_examples=False,
fn=generate_model,
outputs=out_lst,
label="Loose Cloth Exampels"
)
with gr.Column():
overlap_inp = gr.Image(type="filepath", label="Image Normal Overlap")
with gr.Row():
out_final = gr.Model3D(clear_color=[0.0, 0.0, 0.0, 0.0], label="Clothed human")
out_smpl = gr.Model3D(clear_color=[0.0, 0.0, 0.0, 0.0], label="SMPL-X body")
out_final_obj = gr.State()
vis_tensor_path = gr.State()
with gr.Row():
btn_video = gr.Button("Generate Video (~2min)")
with gr.Row():
out_vid = gr.Video(label="Shared on Twitter with #ECON")
# with gr.Row():
# btn_texture = gr.Button("Generate Full-texture")
# with gr.Row():
# prompt = gr.Textbox(
# label="Enter your prompt to texture the mesh",
# max_lines=10,
# placeholder=
# "best quality, extremely detailed, solid color background, super detail, high detail, edge lighting, soft focus, light and dark contrast, 8k, high detail, edge lighting, 3d, c4d, blender, oc renderer, ultra high definition, 3d rendering",
# )
# 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
# )
# progress_text = gr.Text(label='Progress')
# with gr.Tabs():
# with gr.TabItem(label='Images from each viewpoint'):
# viewpoint_images = gr.Gallery(show_label=False)
# with gr.TabItem(label='Result video'):
# result_video = gr.Video(show_label=False)
# with gr.TabItem(label='Output mesh file'):
# output_file = gr.File(show_label=False)
out_lst = [out_smpl, out_final, out_final_obj, overlap_inp, vis_tensor_path]
btn_video.click(
fn=generate_video,
inputs=[vis_tensor_path],
outputs=[out_vid],
)
btn_submit.click(fn=generate_model, inputs=[inp, fitting_step], outputs=out_lst)
# btn_texture.click(
# fn=generate_texture,
# inputs=[out_final_obj, prompt, seed, guidance_scale],
# outputs=[viewpoint_images, result_video, output_file, progress_text]
# )
demo.load(None, None, None, _js=load_js)
if __name__ == "__main__":
# demo.launch(debug=False, enable_queue=False,
# auth=(os.environ['USER'], os.environ['PASSWORD']),
# auth_message="Register at icon.is.tue.mpg.de to get HuggingFace username and password.")
demo.launch(debug=True, enable_queue=True)