lib/load_nsvf.py

import os
import glob
import torch
import numpy as np
import imageio
import json
import torch.nn.functional as F
import cv2


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
    c2w[:,[1,2]] *= -1
    return c2w


def load_nsvf_data(basedir):
    pose_paths = sorted(glob.glob(os.path.join(basedir, 'pose', '*txt')))
    rgb_paths = sorted(glob.glob(os.path.join(basedir, 'rgb', '*png')))

    all_poses = []
    all_imgs = []
    i_split = [[], [], []]
    for i, (pose_path, rgb_path) in enumerate(zip(pose_paths, rgb_paths)):
        i_set = int(os.path.split(rgb_path)[-1][0])
        all_imgs.append((imageio.imread(rgb_path) / 255.).astype(np.float32))
        all_poses.append(np.loadtxt(pose_path).astype(np.float32))
        i_split[i_set].append(i)
    if i_split[2] == []:
        i_split[2] = i_split[1]

    imgs = np.stack(all_imgs, 0)
    poses = np.stack(all_poses, 0)

    H, W = imgs[0].shape[:2]
    with open(os.path.join(basedir, 'intrinsics.txt')) as f:
        focal = float(f.readline().split()[0])

    R = np.sqrt((poses[...,:3,3]**2).sum(-1)).mean()
    render_poses = torch.stack([pose_spherical(angle, -30.0, R) for angle in np.linspace(-180,180,200+1)[:-1]], 0)

    return imgs, poses, render_poses, [H, W, focal], i_split