Text-to-3D
image-to-3d
Chao Xu
sparseneus and elev est
854f0d0
raw
history blame
14 kB
import torch
from torch.utils.data import Dataset
import json
import numpy as np
import os
from PIL import Image
from torchvision import transforms as T
from kornia import create_meshgrid
from models.rays import gen_rays_from_single_image, gen_random_rays_from_single_image
import cv2 as cv
from data.scene import get_boundingbox
def get_ray_directions(H, W, focal, center=None):
"""
Get ray directions for all pixels in camera coordinate.
Reference: https://www.scratchapixel.com/lessons/3d-basic-rendering/
ray-tracing-generating-camera-rays/standard-coordinate-systems
Inputs:
H, W, focal: image height, width and focal length
Outputs:
directions: (H, W, 3), the direction of the rays in camera coordinate
"""
grid = create_meshgrid(H, W, normalized_coordinates=False)[0]
i, j = grid.unbind(-1)
# the direction here is without +0.5 pixel centering as calibration is not so accurate
# see https://github.com/bmild/nerf/issues/24
cent = center if center is not None else [W / 2, H / 2]
directions = torch.stack([(i - cent[0]) / focal[0], (j - cent[1]) / focal[1], torch.ones_like(i)], -1) # (H, W, 3)
return directions
def get_rays(directions, c2w):
"""
Get ray origin and normalized directions in world coordinate for all pixels in one image.
Reference: https://www.scratchapixel.com/lessons/3d-basic-rendering/
ray-tracing-generating-camera-rays/standard-coordinate-systems
Inputs:
directions: (H, W, 3) precomputed ray directions in camera coordinate
c2w: (3, 4) transformation matrix from camera coordinate to world coordinate
Outputs:
rays_o: (H*W, 3), the origin of the rays in world coordinate
rays_d: (H*W, 3), the normalized direction of the rays in world coordinate
"""
# Rotate ray directions from camera coordinate to the world coordinate
rays_d = directions @ c2w[:3, :3].T # (H, W, 3)
# rays_d = rays_d / torch.norm(rays_d, dim=-1, keepdim=True)
# The origin of all rays is the camera origin in world coordinate
rays_o = c2w[:3, 3].expand(rays_d.shape) # (H, W, 3)
rays_d = rays_d.view(-1, 3)
rays_o = rays_o.view(-1, 3)
return rays_o, rays_d
def load_K_Rt_from_P(filename, P=None):
if P is None:
lines = open(filename).read().splitlines()
if len(lines) == 4:
lines = lines[1:]
lines = [[x[0], x[1], x[2], x[3]] for x in (x.split(" ") for x in lines)]
P = np.asarray(lines).astype(np.float32).squeeze()
out = cv.decomposeProjectionMatrix(P)
K = out[0]
R = out[1]
t = out[2]
K = K / K[2, 2]
intrinsics = np.eye(4)
intrinsics[:3, :3] = K
pose = np.eye(4, dtype=np.float32)
pose[:3, :3] = R.transpose() # ? why need transpose here
pose[:3, 3] = (t[:3] / t[3])[:, 0]
return intrinsics, pose # ! return cam2world matrix here
class BlenderDataset(Dataset):
def __init__(self, root_dir, split, scan_id, n_views, train_img_idx=[], test_img_idx=[],
img_wh=[800, 800], clip_wh=[0, 0], original_img_wh=[800, 800],
N_rays=512, h_patch_size=5, near=2.0, far=6.0):
self.root_dir = root_dir
self.split = split
self.img_wh = img_wh
self.clip_wh = clip_wh
self.define_transforms()
self.train_img_idx = train_img_idx
self.test_img_idx = test_img_idx
self.N_rays = N_rays
self.h_patch_size = h_patch_size # used to extract patch for supervision
self.n_views = n_views
self.near, self.far = near, far
self.blender2opencv = np.array([[1, 0, 0, 0], [0, -1, 0, 0], [0, 0, -1, 0], [0, 0, 0, 1]])
with open(os.path.join(self.root_dir, f"transforms_{self.split}.json"), 'r') as f:
self.meta = json.load(f)
self.read_meta(near, far)
# import ipdb; ipdb.set_trace()
self.raw_near_fars = np.stack([np.array([self.near, self.far]) for i in range(len(self.meta['frames']))])
# ! estimate scale_mat
self.scale_mat, self.scale_factor = self.cal_scale_mat(
img_hw=[self.img_wh[1], self.img_wh[0]],
intrinsics=self.all_intrinsics[self.train_img_idx],
extrinsics=self.all_w2cs[self.train_img_idx],
near_fars=self.raw_near_fars[self.train_img_idx],
factor=1.1)
# self.scale_mat = np.eye(4)
# self.scale_factor = 1.0
# import ipdb; ipdb.set_trace()
# * after scaling and translation, unit bounding box
self.scaled_intrinsics, self.scaled_w2cs, self.scaled_c2ws, \
self.scaled_affine_mats, self.scaled_near_fars = self.scale_cam_info()
self.bbox_min = np.array([-1.0, -1.0, -1.0])
self.bbox_max = np.array([1.0, 1.0, 1.0])
self.partial_vol_origin = torch.Tensor([-1., -1., -1.])
self.white_back = True
def read_meta(self, near=2.0, far=6.0):
self.ref_img_idx = self.train_img_idx[0]
ref_c2w = np.array(self.meta['frames'][self.ref_img_idx]['transform_matrix']) @ self.blender2opencv
# ref_c2w = torch.FloatTensor(ref_c2w)
self.ref_c2w = ref_c2w
self.ref_w2c = np.linalg.inv(ref_c2w)
w, h = self.img_wh
self.focal = 0.5 * 800 / np.tan(0.5 * self.meta['camera_angle_x']) # original focal length
self.focal *= self.img_wh[0] / 800 # modify focal length to match size self.img_wh
# bounds, common for all scenes
self.near = near
self.far = far
self.bounds = np.array([self.near, self.far])
# ray directions for all pixels, same for all images (same H, W, focal)
self.directions = get_ray_directions(h, w, [self.focal,self.focal]) # (h, w, 3)
intrinsics = np.eye(4)
intrinsics[:3, :3] = np.array([[self.focal,0,w/2],[0,self.focal,h/2],[0,0,1]]).astype(np.float32)
self.intrinsics = intrinsics
self.image_paths = []
self.poses = []
self.all_rays = []
self.all_images = []
self.all_masks = []
self.all_w2cs = []
self.all_intrinsics = []
for frame in self.meta['frames']:
pose = np.array(frame['transform_matrix']) @ self.blender2opencv
self.poses += [pose]
c2w = torch.FloatTensor(pose)
w2c = np.linalg.inv(c2w)
image_path = os.path.join(self.root_dir, f"{frame['file_path']}.png")
self.image_paths += [image_path]
img = Image.open(image_path)
img = img.resize(self.img_wh, Image.LANCZOS)
img = self.transform(img) # (4, h, w)
self.all_masks += [img[-1:,:]>0]
# img = img[:3, :] * img[ -1:,:] + (1 - img[-1:, :]) # blend A to RGB
img = img[:3, :] * img[ -1:,:]
img = img.numpy() # (3, h, w)
self.all_images += [img]
self.all_masks += []
self.all_intrinsics.append(self.intrinsics)
# - transform from world system to ref-camera system
self.all_w2cs.append(w2c @ np.linalg.inv(self.ref_w2c))
self.all_images = torch.from_numpy(np.stack(self.all_images)).to(torch.float32)
self.all_intrinsics = torch.from_numpy(np.stack(self.all_intrinsics)).to(torch.float32)
self.all_w2cs = torch.from_numpy(np.stack(self.all_w2cs)).to(torch.float32)
# self.img_wh = [self.img_wh[0] - self.clip_wh[0] - self.clip_wh[2],
# self.img_wh[1] - self.clip_wh[1] - self.clip_wh[3]]
def cal_scale_mat(self, img_hw, intrinsics, extrinsics, near_fars, factor=1.):
center, radius, _ = get_boundingbox(img_hw, intrinsics, extrinsics, near_fars)
radius = radius * factor
scale_mat = np.diag([radius, radius, radius, 1.0])
scale_mat[:3, 3] = center.cpu().numpy()
scale_mat = scale_mat.astype(np.float32)
return scale_mat, 1. / radius.cpu().numpy()
def scale_cam_info(self):
new_intrinsics = []
new_near_fars = []
new_w2cs = []
new_c2ws = []
new_affine_mats = []
for idx in range(len(self.all_images)):
intrinsics = self.all_intrinsics[idx]
# import ipdb; ipdb.set_trace()
P = intrinsics @ self.all_w2cs[idx] @ self.scale_mat
P = P.cpu().numpy()[:3, :4]
# - should use load_K_Rt_from_P() to obtain c2w
c2w = load_K_Rt_from_P(None, P)[1]
w2c = np.linalg.inv(c2w)
new_w2cs.append(w2c)
new_c2ws.append(c2w)
new_intrinsics.append(intrinsics)
affine_mat = np.eye(4)
affine_mat[:3, :4] = intrinsics[:3, :3] @ w2c[:3, :4]
new_affine_mats.append(affine_mat)
camera_o = c2w[:3, 3]
dist = np.sqrt(np.sum(camera_o ** 2))
near = dist - 1
far = dist + 1
new_near_fars.append([0.95 * near, 1.05 * far])
new_intrinsics, new_w2cs, new_c2ws, new_affine_mats, new_near_fars = \
np.stack(new_intrinsics), np.stack(new_w2cs), np.stack(new_c2ws), \
np.stack(new_affine_mats), np.stack(new_near_fars)
new_intrinsics = torch.from_numpy(np.float32(new_intrinsics))
new_w2cs = torch.from_numpy(np.float32(new_w2cs))
new_c2ws = torch.from_numpy(np.float32(new_c2ws))
new_affine_mats = torch.from_numpy(np.float32(new_affine_mats))
new_near_fars = torch.from_numpy(np.float32(new_near_fars))
return new_intrinsics, new_w2cs, new_c2ws, new_affine_mats, new_near_fars
def load_poses_all(self, file=f"transforms_train.json"):
with open(os.path.join(self.root_dir, file), 'r') as f:
meta = json.load(f)
c2ws = []
for i,frame in enumerate(meta['frames']):
c2ws.append(np.array(frame['transform_matrix']) @ self.blender2opencv)
return np.stack(c2ws)
def define_transforms(self):
self.transform = T.ToTensor()
def get_conditional_sample(self):
sample = {}
support_idxs = self.train_img_idx
sample['images'] = self.all_images[support_idxs] # (V, 3, H, W)
sample['w2cs'] = self.scaled_w2cs[self.train_img_idx] # (V, 4, 4)
sample['c2ws'] = self.scaled_c2ws[self.train_img_idx] # (V, 4, 4)
sample['near_fars'] = self.scaled_near_fars[self.train_img_idx] # (V, 2)
sample['intrinsics'] = self.scaled_intrinsics[self.train_img_idx][:, :3, :3] # (V, 3, 3)
sample['affine_mats'] = self.scaled_affine_mats[self.train_img_idx] # ! in world space
# sample['scan'] = self.scan_id
sample['scale_factor'] = torch.tensor(self.scale_factor)
sample['scale_mat'] = torch.from_numpy(self.scale_mat)
sample['trans_mat'] = torch.from_numpy(np.linalg.inv(self.ref_w2c))
sample['img_wh'] = torch.from_numpy(np.array(self.img_wh))
sample['partial_vol_origin'] = torch.tensor(self.partial_vol_origin, dtype=torch.float32)
return sample
def __len__(self):
if self.split == 'train':
return self.n_views * 1000
else:
return len(self.test_img_idx) * 1000
def __getitem__(self, idx):
sample = {}
if self.split == 'train':
render_idx = self.train_img_idx[idx % self.n_views]
support_idxs = [idx for idx in self.train_img_idx if idx != render_idx]
else:
# render_idx = idx % self.n_test_images + self.n_train_images
render_idx = self.test_img_idx[idx % len(self.test_img_idx)]
support_idxs = [render_idx]
sample['images'] = self.all_images[support_idxs] # (V, 3, H, W)
sample['w2cs'] = self.scaled_w2cs[support_idxs] # (V, 4, 4)
sample['c2ws'] = self.scaled_c2ws[support_idxs] # (V, 4, 4)
sample['intrinsics'] = self.scaled_intrinsics[support_idxs][:, :3, :3] # (V, 3, 3)
sample['affine_mats'] = self.scaled_affine_mats[support_idxs] # ! in world space
# sample['scan'] = self.scan_id
sample['scale_factor'] = torch.tensor(self.scale_factor)
sample['img_wh'] = torch.from_numpy(np.array(self.img_wh))
sample['partial_vol_origin'] = torch.tensor(self.partial_vol_origin, dtype=torch.float32)
sample['img_index'] = torch.tensor(render_idx)
# - query image
sample['query_image'] = self.all_images[render_idx]
sample['query_c2w'] = self.scaled_c2ws[render_idx]
sample['query_w2c'] = self.scaled_w2cs[render_idx]
sample['query_intrinsic'] = self.scaled_intrinsics[render_idx]
sample['query_near_far'] = self.scaled_near_fars[render_idx]
# sample['meta'] = str(self.scan_id) + "_" + os.path.basename(self.images_list[render_idx])
sample['scale_mat'] = torch.from_numpy(self.scale_mat)
sample['trans_mat'] = torch.from_numpy(np.linalg.inv(self.ref_w2c))
sample['rendering_c2ws'] = self.scaled_c2ws[self.test_img_idx]
sample['rendering_imgs_idx'] = torch.Tensor(np.array(self.test_img_idx).astype(np.int32))
# - generate rays
if self.split == 'val' or self.split == 'test':
sample_rays = gen_rays_from_single_image(
self.img_wh[1], self.img_wh[0],
sample['query_image'],
sample['query_intrinsic'],
sample['query_c2w'],
depth=None,
mask=None)
else:
sample_rays = gen_random_rays_from_single_image(
self.img_wh[1], self.img_wh[0],
self.N_rays,
sample['query_image'],
sample['query_intrinsic'],
sample['query_c2w'],
depth=None,
mask=None,
dilated_mask=None,
importance_sample=False,
h_patch_size=self.h_patch_size
)
sample['rays'] = sample_rays
return sample