app.py

from PIL import Image
import torch
from tqdm.auto import tqdm

from point_e.diffusion.configs import DIFFUSION_CONFIGS, diffusion_from_config
from point_e.diffusion.sampler import PointCloudSampler
from point_e.models.download import load_checkpoint
from point_e.models.configs import MODEL_CONFIGS, model_from_config
from point_e.util.plotting import plot_point_cloud
import streamlit as st

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
st.write('creating base model...')
base_name = 'base40M' # use base300M or base1B for better results
base_model = model_from_config(MODEL_CONFIGS[base_name], device)
base_model.eval()
base_diffusion = diffusion_from_config(DIFFUSION_CONFIGS[base_name])
st.write('creating upsample model...')
upsampler_model = model_from_config(MODEL_CONFIGS['upsample'], device)
upsampler_model.eval()
upsampler_diffusion = diffusion_from_config(DIFFUSION_CONFIGS['upsample'])
st.write('downloading base checkpoint...')
base_model.load_state_dict(load_checkpoint(base_name, device))
st.write('downloading upsampler checkpoint...')
upsampler_model.load_state_dict(load_checkpoint('upsample', device))