3DFuse / pc_project.py
Hoodady's picture
Duplicate from jyseo/3DFuse
19a1abb
import os
import numpy as np
import torch
import torch.nn as nn
from PIL import Image
from my.utils import tqdm
from pytorch3d.structures import Pointclouds
from pytorch3d.renderer.cameras import PerspectiveCameras
from pytorch3d.renderer import (
PointsRasterizer,
AlphaCompositor,
look_at_view_transform,
)
import torch.nn.functional as F
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
class PointsRenderer(nn.Module):
"""
Modified version of Pytorch3D PointsRenderer
"""
def __init__(self, rasterizer, compositor) -> None:
super().__init__()
self.rasterizer = rasterizer
self.compositor = compositor
def to(self, device):
# Manually move to device rasterizer as the cameras
# within the class are not of type nn.Module
self.rasterizer = self.rasterizer.to(device)
self.compositor = self.compositor.to(device)
return self
def forward(self, point_clouds, **kwargs) -> torch.Tensor:
fragments = self.rasterizer(point_clouds, **kwargs)
# import pdb; pdb.set_trace()
depth_map = fragments[1][0,...,:1]
# Construct weights based on the distance of a point to the true point.
# However, this could be done differently: e.g. predicted as opposed
# to a function of the weights.
r = self.rasterizer.raster_settings.radius
dists2 = fragments.dists.permute(0, 3, 1, 2)
weights = 1 - dists2 / (r * r)
images = self.compositor(
fragments.idx.long().permute(0, 3, 1, 2),
weights,
point_clouds.features_packed().permute(1, 0),
**kwargs,
)
# permute so image comes at the end
images = images.permute(0, 2, 3, 1)
return images, depth_map
def render_depth_from_cloud(points, angles, raster_settings, device,calibration_value=0):
radius = 2.3
horizontal = angles[0]+calibration_value
elevation = angles[1]
FoV = angles[2]
camera = py3d_camera(radius, elevation, horizontal, FoV, device)
point_loc = torch.tensor(points.coords).to(device)
colors = torch.tensor(np.stack([points.channels["R"], points.channels["G"], points.channels["B"]], axis=-1)).to(device)
matching_rotation = torch.tensor([[[1.0, 0.0, 0.0],
[0.0, 0.0, 1.0],
[0.0, -1.0, 0.0]]]).to(device)
rot_points = (matching_rotation @ point_loc[...,None]).squeeze()
point_cloud = Pointclouds(points=[rot_points], features=[colors])
_, raw_depth_map = pointcloud_renderer(point_cloud, camera, raster_settings, device)
disparity = camera.focal_length[0,0] / (raw_depth_map + 1e-9)
max_disp = torch.max(disparity)
min_disp = torch.min(disparity[disparity > 0])
norm_disparity = (disparity - min_disp) / (max_disp - min_disp)
mask = norm_disparity > 0
norm_disparity = norm_disparity * mask
depth_map = F.interpolate(norm_disparity.permute(2,0,1)[None,...],size=512,mode='bilinear')[0]
depth_map = depth_map.repeat(3,1,1)
return depth_map
def py3d_camera(radius, elevation, horizontal, FoV, device, img_size=800):
fov_rad = torch.deg2rad(torch.tensor(FoV))
focal = 1 / torch.tan(fov_rad / 2) * (2. / 2)
focal_length = torch.tensor([[focal,focal]]).float()
image_size = torch.tensor([[img_size,img_size]]).double()
R, T = look_at_view_transform(dist=radius, elev=elevation, azim=horizontal, degrees=True)
camera = PerspectiveCameras(
R=R,
T=T,
focal_length=focal_length,
image_size=image_size,
device=device,
)
return camera
def pointcloud_renderer(point_cloud, camera, raster_settings, device):
camera = camera.to(device)
rasterizer = PointsRasterizer(cameras=camera, raster_settings=raster_settings)
renderer = PointsRenderer(
rasterizer=rasterizer,
compositor=AlphaCompositor()
).to(device)
image = renderer(point_cloud)
return image
def point_e(device,exp_dir):
print('creating base model...')
base_name = 'base1B' # 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])
print('creating upsample model...')
upsampler_model = model_from_config(MODEL_CONFIGS['upsample'], device)
upsampler_model.eval()
upsampler_diffusion = diffusion_from_config(DIFFUSION_CONFIGS['upsample'])
print('downloading base checkpoint...')
base_model.load_state_dict(load_checkpoint(base_name, device))
print('downloading upsampler checkpoint...')
upsampler_model.load_state_dict(load_checkpoint('upsample', device))
sampler = PointCloudSampler(
device=device,
models=[base_model, upsampler_model],
diffusions=[base_diffusion, upsampler_diffusion],
num_points=[1024, 4096 - 1024],
aux_channels=['R', 'G', 'B'],
guidance_scale=[3.0, 3.0],
)
img = Image.open(os.path.join(exp_dir,'initial_image','instance0.png'))
samples = None
for x in tqdm(sampler.sample_batch_progressive(batch_size=1, model_kwargs=dict(images=[img]))):
samples = x
pc = sampler.output_to_point_clouds(samples)[0]
return pc
def point_e_gradio(img,device):
print('creating base model...')
base_name = 'base1B' # 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])
print('creating upsample model...')
upsampler_model = model_from_config(MODEL_CONFIGS['upsample'], device)
upsampler_model.eval()
upsampler_diffusion = diffusion_from_config(DIFFUSION_CONFIGS['upsample'])
print('downloading base checkpoint...')
base_model.load_state_dict(load_checkpoint(base_name, device))
print('downloading upsampler checkpoint...')
upsampler_model.load_state_dict(load_checkpoint('upsample', device))
sampler = PointCloudSampler(
device=device,
models=[base_model, upsampler_model],
diffusions=[base_diffusion, upsampler_diffusion],
num_points=[1024, 4096 - 1024],
aux_channels=['R', 'G', 'B'],
guidance_scale=[3.0, 3.0],
)
samples = None
for x in tqdm(sampler.sample_batch_progressive(batch_size=1, model_kwargs=dict(images=[img]))):
samples = x
pc = sampler.output_to_point_clouds(samples)[0]
return pc