4DGaussians/scene/dataset_readers.py
guanjunwu 8bf73f413d 123
2023-09-24 19:51:57 +08:00

486 lines
19 KiB
Python

#
# Copyright (C) 2023, Inria
# GRAPHDECO research group, https://team.inria.fr/graphdeco
# All rights reserved.
#
# This software is free for non-commercial, research and evaluation use
# under the terms of the LICENSE.md file.
#
# For inquiries contact george.drettakis@inria.fr
#
import os
import sys
from PIL import Image
from typing import NamedTuple
from scene.colmap_loader import read_extrinsics_text, read_intrinsics_text, qvec2rotmat, \
read_extrinsics_binary, read_intrinsics_binary, read_points3D_binary, read_points3D_text
from scene.hyper_loader import Load_hyper_data, format_hyper_data
import torchvision.transforms as transforms
from utils.graphics_utils import getWorld2View2, focal2fov, fov2focal
import numpy as np
import torch
import json
from pathlib import Path
from plyfile import PlyData, PlyElement
from utils.sh_utils import SH2RGB
from scene.gaussian_model import BasicPointCloud
from tqdm import tqdm
class CameraInfo(NamedTuple):
uid: int
R: np.array
T: np.array
FovY: np.array
FovX: np.array
image: np.array
image_path: str
image_name: str
width: int
height: int
time : float
class SceneInfo(NamedTuple):
point_cloud: BasicPointCloud
train_cameras: list
test_cameras: list
video_cameras: list
nerf_normalization: dict
ply_path: str
maxtime: int
def getNerfppNorm(cam_info):
def get_center_and_diag(cam_centers):
cam_centers = np.hstack(cam_centers)
avg_cam_center = np.mean(cam_centers, axis=1, keepdims=True)
center = avg_cam_center
dist = np.linalg.norm(cam_centers - center, axis=0, keepdims=True)
diagonal = np.max(dist)
return center.flatten(), diagonal
cam_centers = []
for cam in cam_info:
W2C = getWorld2View2(cam.R, cam.T)
C2W = np.linalg.inv(W2C)
cam_centers.append(C2W[:3, 3:4])
center, diagonal = get_center_and_diag(cam_centers)
radius = diagonal * 1.1
translate = -center
return {"translate": translate, "radius": radius}
def readColmapCameras(cam_extrinsics, cam_intrinsics, images_folder):
cam_infos = []
for idx, key in enumerate(cam_extrinsics):
sys.stdout.write('\r')
# the exact output you're looking for:
sys.stdout.write("Reading camera {}/{}".format(idx+1, len(cam_extrinsics)))
sys.stdout.flush()
extr = cam_extrinsics[key]
intr = cam_intrinsics[extr.camera_id]
height = intr.height
width = intr.width
uid = intr.id
R = np.transpose(qvec2rotmat(extr.qvec))
T = np.array(extr.tvec)
if intr.model in ["SIMPLE_PINHOLE", "SIMPLE_RADIAL"]:
focal_length_x = intr.params[0]
FovY = focal2fov(focal_length_x, height)
FovX = focal2fov(focal_length_x, width)
elif intr.model=="PINHOLE":
focal_length_x = intr.params[0]
focal_length_y = intr.params[1]
FovY = focal2fov(focal_length_y, height)
FovX = focal2fov(focal_length_x, width)
else:
assert False, "Colmap camera model not handled: only undistorted datasets (PINHOLE or SIMPLE_PINHOLE cameras) supported!"
image_path = os.path.join(images_folder, os.path.basename(extr.name))
image_name = os.path.basename(image_path).split(".")[0]
image = Image.open(image_path)
cam_info = CameraInfo(uid=uid, R=R, T=T, FovY=FovY, FovX=FovX, image=image,
image_path=image_path, image_name=image_name, width=width, height=height,
time = 0)
cam_infos.append(cam_info)
sys.stdout.write('\n')
return cam_infos
def fetchPly(path):
plydata = PlyData.read(path)
vertices = plydata['vertex']
positions = np.vstack([vertices['x'], vertices['y'], vertices['z']]).T
colors = np.vstack([vertices['red'], vertices['green'], vertices['blue']]).T / 255.0
normals = np.vstack([vertices['nx'], vertices['ny'], vertices['nz']]).T
return BasicPointCloud(points=positions, colors=colors, normals=normals)
def storePly(path, xyz, rgb):
# Define the dtype for the structured array
dtype = [('x', 'f4'), ('y', 'f4'), ('z', 'f4'),
('nx', 'f4'), ('ny', 'f4'), ('nz', 'f4'),
('red', 'u1'), ('green', 'u1'), ('blue', 'u1')]
normals = np.zeros_like(xyz)
elements = np.empty(xyz.shape[0], dtype=dtype)
attributes = np.concatenate((xyz, normals, rgb), axis=1)
elements[:] = list(map(tuple, attributes))
# Create the PlyData object and write to file
vertex_element = PlyElement.describe(elements, 'vertex')
ply_data = PlyData([vertex_element])
ply_data.write(path)
def readColmapSceneInfo(path, images, eval, llffhold=8):
try:
cameras_extrinsic_file = os.path.join(path, "sparse/0", "images.bin")
cameras_intrinsic_file = os.path.join(path, "sparse/0", "cameras.bin")
cam_extrinsics = read_extrinsics_binary(cameras_extrinsic_file)
cam_intrinsics = read_intrinsics_binary(cameras_intrinsic_file)
except:
cameras_extrinsic_file = os.path.join(path, "sparse/0", "images.txt")
cameras_intrinsic_file = os.path.join(path, "sparse/0", "cameras.txt")
cam_extrinsics = read_extrinsics_text(cameras_extrinsic_file)
cam_intrinsics = read_intrinsics_text(cameras_intrinsic_file)
reading_dir = "images" if images == None else images
cam_infos_unsorted = readColmapCameras(cam_extrinsics=cam_extrinsics, cam_intrinsics=cam_intrinsics, images_folder=os.path.join(path, reading_dir))
cam_infos = sorted(cam_infos_unsorted.copy(), key = lambda x : x.image_name)
if eval:
train_cam_infos = [c for idx, c in enumerate(cam_infos) if idx % llffhold != 0]
test_cam_infos = [c for idx, c in enumerate(cam_infos) if idx % llffhold == 0]
else:
train_cam_infos = cam_infos
test_cam_infos = []
nerf_normalization = getNerfppNorm(train_cam_infos)
ply_path = os.path.join(path, "sparse/0/points3D.ply")
bin_path = os.path.join(path, "sparse/0/points3D.bin")
txt_path = os.path.join(path, "sparse/0/points3D.txt")
if not os.path.exists(ply_path):
print("Converting point3d.bin to .ply, will happen only the first time you open the scene.")
try:
xyz, rgb, _ = read_points3D_binary(bin_path)
except:
xyz, rgb, _ = read_points3D_text(txt_path)
storePly(ply_path, xyz, rgb)
try:
pcd = fetchPly(ply_path)
except:
pcd = None
scene_info = SceneInfo(point_cloud=pcd,
train_cameras=train_cam_infos,
test_cameras=test_cam_infos,
video_cameras=train_cam_infos,
maxtime=0,
nerf_normalization=nerf_normalization,
ply_path=ply_path)
return scene_info
def generateCamerasFromTransforms(path, template_transformsfile, extension, maxtime):
trans_t = lambda t : torch.Tensor([
[1,0,0,0],
[0,1,0,0],
[0,0,1,t],
[0,0,0,1]]).float()
rot_phi = lambda phi : torch.Tensor([
[1,0,0,0],
[0,np.cos(phi),-np.sin(phi),0],
[0,np.sin(phi), np.cos(phi),0],
[0,0,0,1]]).float()
rot_theta = lambda th : torch.Tensor([
[np.cos(th),0,-np.sin(th),0],
[0,1,0,0],
[np.sin(th),0, np.cos(th),0],
[0,0,0,1]]).float()
def pose_spherical(theta, phi, radius):
c2w = trans_t(radius)
c2w = rot_phi(phi/180.*np.pi) @ c2w
c2w = rot_theta(theta/180.*np.pi) @ c2w
c2w = torch.Tensor(np.array([[-1,0,0,0],[0,0,1,0],[0,1,0,0],[0,0,0,1]])) @ c2w
return c2w
cam_infos = []
# generate render poses and times
render_poses = torch.stack([pose_spherical(angle, -30.0, 4.0) for angle in np.linspace(-180,180,40+1)[:-1]], 0)
render_times = torch.linspace(0,maxtime,render_poses.shape[0])
with open(os.path.join(path, template_transformsfile)) as json_file:
template_json = json.load(json_file)
fovx = template_json["camera_angle_x"]
# load a single image to get image info.
for idx, frame in enumerate(template_json["frames"]):
cam_name = os.path.join(path, frame["file_path"] + extension)
image_path = os.path.join(path, cam_name)
image_name = Path(cam_name).stem
image = Image.open(image_path)
im_data = np.array(image.convert("RGBA"))
break
# format infomation
for idx, (time, poses) in enumerate(zip(render_times,render_poses)):
time = time/maxtime
matrix = np.linalg.inv(np.array(poses))
R = -np.transpose(matrix[:3,:3])
R[:,0] = -R[:,0]
T = -matrix[:3, 3]
fovy = focal2fov(fov2focal(fovx, image.size[0]), image.size[1])
FovY = fovy
FovX = fovx
cam_infos.append(CameraInfo(uid=idx, R=R, T=T, FovY=FovY, FovX=FovX, image=image,
image_path=None, image_name=None, width=image.size[0], height=image.size[1],
time = time))
return cam_infos
def readCamerasFromTransforms(path, transformsfile, white_background, extension=".png", mapper = {}):
cam_infos = []
with open(os.path.join(path, transformsfile)) as json_file:
contents = json.load(json_file)
fovx = contents["camera_angle_x"]
frames = contents["frames"]
for idx, frame in enumerate(frames):
cam_name = os.path.join(path, frame["file_path"] + extension)
time = mapper[frame["time"]]
matrix = np.linalg.inv(np.array(frame["transform_matrix"]))
R = -np.transpose(matrix[:3,:3])
R[:,0] = -R[:,0]
T = -matrix[:3, 3]
image_path = os.path.join(path, cam_name)
image_name = Path(cam_name).stem
image = Image.open(image_path)
im_data = np.array(image.convert("RGBA"))
bg = np.array([1,1,1]) if white_background else np.array([0, 0, 0])
norm_data = im_data / 255.0
arr = norm_data[:,:,:3] * norm_data[:, :, 3:4] + bg * (1 - norm_data[:, :, 3:4])
image = Image.fromarray(np.array(arr*255.0, dtype=np.byte), "RGB")
fovy = focal2fov(fov2focal(fovx, image.size[0]), image.size[1])
FovY = fovy
FovX = fovx
cam_infos.append(CameraInfo(uid=idx, R=R, T=T, FovY=FovY, FovX=FovX, image=image,
image_path=image_path, image_name=image_name, width=image.size[0], height=image.size[1],
time = time))
return cam_infos
def read_timeline(path):
with open(os.path.join(path, "transforms_train.json")) as json_file:
train_json = json.load(json_file)
with open(os.path.join(path, "transforms_test.json")) as json_file:
test_json = json.load(json_file)
time_line = [frame["time"] for frame in train_json["frames"]] + [frame["time"] for frame in test_json["frames"]]
time_line = set(time_line)
time_line = list(time_line)
time_line.sort()
timestamp_mapper = {}
max_time_float = max(time_line)
for index, time in enumerate(time_line):
# timestamp_mapper[time] = index
timestamp_mapper[time] = time/max_time_float
return timestamp_mapper, max_time_float
def readNerfSyntheticInfo(path, white_background, eval, extension=".png"):
timestamp_mapper, max_time = read_timeline(path)
print("Reading Training Transforms")
train_cam_infos = readCamerasFromTransforms(path, "transforms_train.json", white_background, extension, timestamp_mapper)
print("Reading Test Transforms")
test_cam_infos = readCamerasFromTransforms(path, "transforms_test.json", white_background, extension, timestamp_mapper)
print("Generating Video Transforms")
video_cam_infos = generateCamerasFromTransforms(path, "transforms_train.json", extension, max_time)
if not eval:
train_cam_infos.extend(test_cam_infos)
test_cam_infos = []
nerf_normalization = getNerfppNorm(train_cam_infos)
ply_path = os.path.join(path, "points3d.ply")
if not os.path.exists(ply_path):
# Since this data set has no colmap data, we start with random points
num_pts = 100_000
print(f"Generating random point cloud ({num_pts})...")
# We create random points inside the bounds of the synthetic Blender scenes
xyz = np.random.random((num_pts, 3)) * 2.6 - 1.3
shs = np.random.random((num_pts, 3)) / 255.0
pcd = BasicPointCloud(points=xyz, colors=SH2RGB(shs), normals=np.zeros((num_pts, 3)))
storePly(ply_path, xyz, SH2RGB(shs) * 255)
try:
pcd = fetchPly(ply_path)
except:
pcd = None
scene_info = SceneInfo(point_cloud=pcd,
train_cameras=train_cam_infos,
test_cameras=test_cam_infos,
video_cameras=video_cam_infos,
nerf_normalization=nerf_normalization,
ply_path=ply_path,
maxtime=max_time
)
return scene_info
def format_infos(dataset,split):
# loading
cameras = []
image = dataset[0][0]
if split == "train":
for idx in tqdm(range(len(dataset))):
image_path = None
image_name = f"{idx}"
time = dataset.image_times[idx]
# matrix = np.linalg.inv(np.array(pose))
R,T = dataset.load_pose(idx)
FovX = focal2fov(dataset.focal[0], image.shape[1])
FovY = focal2fov(dataset.focal[0], image.shape[2])
cameras.append(CameraInfo(uid=idx, R=R, T=T, FovY=FovY, FovX=FovX, image=image,
image_path=image_path, image_name=image_name, width=image.shape[2], height=image.shape[1],
time = time))
return cameras
def readHyperDataInfos(datadir,use_bg_points,eval):
data_class = Load_hyper_data(datadir,0.25,use_bg_points,add_cam=True)
train_cam_infos = format_hyper_data(data_class,"train")
test_cam_infos = format_hyper_data(data_class,"test")
max_time = data_class.max_time
video_cam_infos = format_render_poses(train_cam_infos.val_poses, train_cam_infos)
ply_path = os.path.join(datadir, "points3d.ply")
# ply_path = os.path.join(datadir, "points.npy")
# if not os.path.exists(ply_path):
# Since this data set has no colmap data, we start with random points
num_pts = 100_000
print(f"Generating random point cloud ({num_pts})...")
# from scene.ray_utils import compute_bbox_by_cam_frustrm_hyper
# We create random points inside the bounds of the synthetic Blender scenes
# xyz_min, xyz_max = compute_bbox_by_cam_frustrm_hyper(data_class)
bounds = 10
xyz_min = np.array([-bounds,-bounds,-bounds])
xyz_max = np.array([bounds,bounds,bounds])
print("scene bounding box:",xyz_min, xyz_max)
center = (xyz_min + xyz_max)/2
xyz = (np.random.random((num_pts, 3)))* (np.array(xyz_max) - np.array(xyz_min)) + np.array(center)
shs = np.random.random((xyz.shape[0], 3)) / 255.0
pcd = BasicPointCloud(points=xyz, colors=SH2RGB(shs), normals=np.zeros((xyz.shape[0], 3)))
storePly(ply_path, xyz, SH2RGB(shs) * 255)
# try:
#
# pcd = fetchPly(ply_path)
# except:
# pcd = None
nerf_normalization = getNerfppNorm(train_cam_infos)
scene_info = SceneInfo(point_cloud=pcd,
train_cameras=train_cam_infos,
test_cameras=test_cam_infos,
video_cameras=video_cam_infos,
nerf_normalization=nerf_normalization,
ply_path=ply_path,
maxtime=max_time
)
return scene_info
def format_render_poses(poses,data_infos):
cameras = []
tensor_to_pil = transforms.ToPILImage()
len_poses = len(poses)
times = [i/len_poses for i in range(len_poses)]
image = data_infos[0][0]
for idx, p in tqdm(enumerate(poses)):
# image = None
image_path = None
image_name = f"{idx}"
time = times[idx]
pose = np.eye(4)
pose[:3,:] = p[:3,:]
# matrix = np.linalg.inv(np.array(pose))
R = pose[:3,:3]
R = - R
R[:,0] = -R[:,0]
T = -pose[:3,3].dot(R)
FovX = focal2fov(data_infos.focal[0], image.shape[2])
FovY = focal2fov(data_infos.focal[0], image.shape[1])
cameras.append(CameraInfo(uid=idx, R=R, T=T, FovY=FovY, FovX=FovX, image=image,
image_path=image_path, image_name=image_name, width=image.shape[2], height=image.shape[1],
time = time))
return cameras
def read3DVideoInfo(datadir,use_bg_points,eval):
# loading all the data follow hexplane format
ply_path = os.path.join(datadir, "points3d.ply")
from scene.neural_3D_dataset_NDC import Neural3D_NDC_Dataset
train_dataset = Neural3D_NDC_Dataset(
datadir,
"train",
2.0,
time_scale=1,
scene_bbox_min=[-2.5, -2.0, -1.0],
scene_bbox_max=[2.5, 2.0, 1.0],
eval_index=0,
)
test_dataset = Neural3D_NDC_Dataset(
datadir,
"test",
2.0,
time_scale=1,
scene_bbox_min=[-2.5, -2.0, -1.0],
scene_bbox_max=[2.5, 2.0, 1.0],
eval_index=0,
)
train_cam_infos = format_infos(train_dataset,"train")
# test_cam_infos = format_infos(test_dataset,"test")
val_cam_infos = format_render_poses(test_dataset.val_poses,test_dataset)
nerf_normalization = getNerfppNorm(train_cam_infos)
# create pcd
# if not os.path.exists(ply_path):
# Since this data set has no colmap data, we start with random points
num_pts = 2000
print(f"Generating random point cloud ({num_pts})...")
threshold = 2.5
xyz_max = np.array([threshold, threshold, threshold])
xyz_min = np.array([-threshold, -threshold, -threshold])
center = (xyz_max - xyz_min)/2
# We create random points inside the bounds of the synthetic Blender scenes
xyz = np.random.random((num_pts, 3)) * (xyz_max-xyz_min) -center
print("point cloud initialization:",xyz.max(axis=0),xyz.min(axis=0))
shs = np.random.random((num_pts, 3)) / 255.0
pcd = BasicPointCloud(points=xyz, colors=SH2RGB(shs), normals=np.zeros((num_pts, 3)))
storePly(ply_path, xyz, SH2RGB(shs) * 255)
try:
# xyz = np.load
pcd = fetchPly(ply_path)
except:
pcd = None
scene_info = SceneInfo(point_cloud=pcd,
train_cameras=train_dataset,
test_cameras=test_dataset,
video_cameras=val_cam_infos,
nerf_normalization=nerf_normalization,
ply_path=ply_path,
maxtime=300
)
return scene_info
sceneLoadTypeCallbacks = {
"Colmap": readColmapSceneInfo,
"Blender" : readNerfSyntheticInfo,
"hyper" : readHyperDataInfos,
"3dvideo" : read3DVideoInfo,
}