Shap-E 3D Generation

Two weeks ago, OpenAI open-sourced Shap-E for 3D generation. The paper address is: https://arxiv.org/pdf/2305.02463.pdf

Today, I will try it out using Colab:

Installation

!git clone https://github.com/openai/shap-e.git
%cd shap-e
!pip install -e .

Text-3D

Let's first generate a small car to see the effect.

import torch
from shap_e.diffusion.sample import sample_latents
from shap_e.diffusion.gaussian_diffusion import diffusion_from_config
from shap_e.models.download import load_model, load_config
from shap_e.util.notebooks import create_pan_cameras, decode_latent_images, gif_widget

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
xm = load_model('transmitter', device=device)
model = load_model('text300M', device=device)
diffusion = diffusion_from_config(load_config('diffusion'))

batch_size = 4
guidance_scale = 15.0
prompt = "a car"

latents = sample_latents(
    batch_size=batch_size,
    model=model,
    diffusion=diffusion,
    guidance_scale=guidance_scale,
    model_kwargs=dict(texts=[prompt] * batch_size),
    progress=True,
    clip_denoised=True,
    use_fp16=True,
    use_karras=True,
    karras_steps=64,
    sigma_min=1e-3,
    sigma_max=160,
    s_churn=0,
)

render_mode = 'nerf'  # you can change this to 'stf'
size = 64  # this is the size of the renders; higher values take longer to render.

cameras = create_pan_cameras(size, device)
for i, latent in enumerate(latents):
    images = decode_latent_images(xm, latent, cameras, rendering_mode=render_mode)
    display(gif_widget(images))

Of course, you can also generate other things, you can refer to the various samples provided by the official source.

https://github.com/openai/shap-e/blob/main/samples.md

The generated 3D model can be downloaded and saved in .ply format by running this script:

from shap_e.util.notebooks import decode_latent_mesh

for i, latent in enumerate(latents):
    with open(f'example_mesh_{i}.ply', 'wb') as f:
        decode_latent_mesh(xm, latent).tri_mesh().write_ply(f)

Afterwards, the .ply file will be stored under the shap-e folder, right-click to download and save it locally.

Image-3D

If you have an image, Shap-E can help you turn it into a 3D model.

Upload the image to the shap-e folder.

Run the script:

import torch

from shap_e.diffusion.sample import sample_latents
from shap_e.diffusion.gaussian_diffusion import diffusion_from_config
from shap_e.models.download import load_model, load_config
from shap_e.util.notebooks import create_pan_cameras, decode_latent_images, gif_widget
from shap_e.util.image_util import load_image

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

xm = load_model('transmitter', device=device)
model = load_model('image300M', device=device)
diffusion = diffusion_from_config(load_config('diffusion'))

batch_size = 4
guidance_scale = 3.0

image = load_image("car.png")

latents = sample_latents(
    batch_size=batch_size,
    model=model,
    diffusion=diffusion,
    guidance_scale=guidance_scale,
    model_kwargs=dict(images=[image] * batch_size),
    progress=True,
    clip_denoised=True,
    use_fp16=True,
    use_karras=True,
    karras_steps=64,
    sigma_min=1e-3,
    sigma_max=160,
    s_churn=0,
)

render_mode = 'nerf' # you can change this to 'stf' for mesh rendering
size = 64 # this is the size of the renders; higher values take longer to render.

cameras = create_pan_cameras(size, device)
for i, latent in enumerate(latents):
    images = decode_latent_images(xm, latent, cameras, rendering_mode=render_mode)
    display(gif_widget(images))

Thus, the 3D model based on this image is generated.

Although it's ugly enough to make one cry, we believe that the wisdom of the open-source community will gradually iterate this AI tool.

It feels like all the work is going to be taken over by AI, so what should us humans do?👇