thewhole's picture
Update app.py
cb5fbf2 verified
import gradio as gr
import numpy as np
import os
import subprocess
from datetime import datetime
os.system('git clone https://huggingface.co./camenduru/GaussianDreamer')
os.system('pip install ./gaussiansplatting/submodules/diff-gaussian-rasterization')
os.system('pip install ./GaussianDreamer/nerfacc-0.5.3-cp310-cp310-linux_x86_64.whl')
os.system('pip install ./gaussiansplatting/submodules/simple-knn')
# os.system('pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch')
os.system('pip install -q https://huggingface.co./camenduru/CoDeF/resolve/main/tinycudann-1.7-cp310-cp310-linux_x86_64.whl')
os.system('git clone https://github.com/openai/shap-e.git')
os.system('pip install -e ./shap-e')
os.system('mv ./GaussianDreamer/shapE_finetuned_with_330kdata.pth ./load/shapE_finetuned_with_330kdata.pth')
example_inputs = [[
"A fox."
], [
"fries and a hamburger."
], [
"Viking axe, fantasy, weapon, blender, 8k, HD."
], [
"ferrari convertible, trending on artstation, ultra realistic, 4k, HD"
], [
"flamethrower, with fire, scifi, cyberpunk, photorealistic, 8K, HD"
], [
"Blue and white porcelain Viking axe"
], [
"a DSLR photo of a small saguaro cactus planted in a clay pot"
], [
"a zoomed out DSLR photo of an amigurumi motorcycle"
], [
"a DSLR photo of a teapot shaped like an elephant head where its snout acts as the spout"
], [
"a DSLR photo of a wine bottle and full wine glass on a chessboard"
], [
"a panda wearing a necktie and sitting in an office chair"
], [
"a spanish galleon sailing on the open sea"
], [
"airplane, fighter, steampunk style, ultra realistic, 4k, HD"
]]
example_outputs_1 = [
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/a_fox.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/fries_and_a_hamburger.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/Viking_axe,_fantasy,_weapon,_blender,_8k,_HD.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/ferrari_convertible,_trending_on_artstation,_ultra_realistic,_4k,_HD.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/flamethrower,_with_fire,_scifi,_cyberpunk,_photorealistic,_8K,_HD.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/Blue_and_white_porcelain_Viking_axe.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/a_DSLR_photo_of_a_small_saguaro_cactus_planted_in_a_clay_pot.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/a_zoomed_out_DSLR_photo_of_an_amigurumi_motorcycle.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/a_DSLR_photo_of_a_teapot_shaped_like_an_elephant_head_where_its_snout_acts_as_the_spout.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/a_DSLR_photo_of_a_wine_bottle_and_full_wine_glass_on_a_chessboard.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/a_panda_wearing_a_necktie_and_sitting_in_an_office_chair.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/a_spanish_galleon_sailing_on_the_open_sea.mp4'), autoplay=True),
gr.Video(value=os.path.join(os.path.dirname(__file__), 'example/airplane,_fighter,_steampunk_style,_ultra_realistic,_4k,_HD.mp4'), autoplay=True)
]
subprocess.run([
f'python shape.py'],
shell=True)
def main(prompt, iteration,CFG, seed):
if [prompt] in example_inputs:
return example_outputs_1[example_inputs.index([prompt])]
seed = int(seed)
iteration = int(iteration)
print('==> User Prompt:', prompt)
timestamp = datetime.now().strftime("%Y%m%d-%H%M%S")
print('==> User Shell:', f'python launch.py --config configs/gaussiandreamer-sd.yaml --train --gpu 0 system.prompt_processor.prompt="{prompt}" seed={seed} system.guidance.guidance_scale={CFG} trainer.max_steps={iteration} use_timestamp=False timestamp="{timestamp}" ',)
subprocess.run([
f'python launch.py --config configs/gaussiandreamer-sd.yaml --train --gpu 0 system.prompt_processor.prompt="{prompt}" seed={seed} system.guidance.guidance_scale={CFG} trainer.max_steps={iteration} use_timestamp=False timestamp="{timestamp}" '],
shell=True)
path= os.path.join("./outputs/gaussiandreamer-sd",f'{prompt.replace(" ","_")}{timestamp}',f"save/it{iteration}-test.mp4")
print('==> Save path:', path)
return gr.Video(value=path, autoplay=True)
with gr.Blocks() as demo:
gr.Markdown("# <center>GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models</center>")
gr.Markdown("This live demo allows you to generate high-quality 3D content using text prompts. The outputs are 360° rendered 3d video.<br> \
It is based on Stable Diffusion 2.1-base. Please check out our <strong><a href=https://taoranyi.com/gaussiandreamer/>Project Page</a> / <a href=https://arxiv.org/abs/2310.08529>Paper</a> / <a href=https://github.com/hustvl/GaussianDreamer>Code</a></strong> if you want to learn more about our method!<br> \
Note that this demo is running on T4, the running time might be longer than the reported 15 minutes (1200 iterations) on RTx 3090.<br> \
&copy; This Gradio space is developed by <a href=https://taoranyi.com/>Taoran Yi</a>.")
gr.Interface(fn=main, inputs=[gr.Textbox(lines=2, value="fries and a hamburger.", label="Your prompt"),
gr.Slider(0, 2000, value=900, label="Number of iteration (using 1200 in paper)"),
gr.Slider(80, 200, value=100, label="CFG"),
gr.Number(value=0, label="Seed")],
outputs=["playable_video"],
examples=example_inputs,
cache_examples=True,
concurrency_limit=1)
demo.launch()