load_llff.py

import numpy as np
import torch
import os, imageio


########## Slightly modified version of LLFF data loading code 
##########  see https://github.com/Fyusion/LLFF for original

########## Adapted to DaVinci endoscopic surgery datasets

def _minify(basedir, factors=[], dir_name='images', resolutions=[]):
    needtoload = False
    for r in factors:
        imgdir = os.path.join(basedir, '{}_{}'.format(dir_name, r))
        if not os.path.exists(imgdir):
            needtoload = True
    for r in resolutions:
        imgdir = os.path.join(basedir, '{}_{}x{}'.format(dir_name, r[1], r[0]))
        if not os.path.exists(imgdir):
            needtoload = True
    if not needtoload:
        return
    
    from shutil import copy
    from subprocess import check_output
    
    imgdir = os.path.join(basedir, dir_name)
    imgs = [os.path.join(imgdir, f) for f in sorted(os.listdir(imgdir))]
    imgs = [f for f in imgs if any([f.endswith(ex) for ex in ['JPG', 'jpg', 'png', 'jpeg', 'PNG']])]
    imgdir_orig = imgdir
    
    wd = os.getcwd()

    for r in factors + resolutions:
        if isinstance(r, int):
            name = '{}_{}'.format(dir_name, r)
            resizearg = '{}%'.format(100./r)
        else:
            name = '{}_{}x{}'.format(dir_name, r[1], r[0])
            resizearg = '{}x{}'.format(r[1], r[0])
        imgdir = os.path.join(basedir, name)
        if os.path.exists(imgdir):
            continue
            
        print('Minifying', r, basedir)
        
        os.makedirs(imgdir)
        check_output('cp {}/* {}'.format(imgdir_orig, imgdir), shell=True)
        
        ext = imgs[0].split('.')[-1]
        args = ' '.join(['mogrify', '-resize', resizearg, '-format', 'png', '*.{}'.format(ext)])
        print(args)
        os.chdir(imgdir)
        check_output(args, shell=True)
        os.chdir(wd)
        
        if ext != 'png':
            check_output('rm {}/*.{}'.format(imgdir, ext), shell=True)
            print('Removed duplicates')
        print('Done')
            


def _preprocess_imgs(basedir, dir_name='images', factor=None, width=None, height=None, check_fn=lambda x: True):
    img0 = [os.path.join(basedir, dir_name, f) for f in sorted(os.listdir(os.path.join(basedir, dir_name))) if check_fn(f, 0)][0]
    sh = imageio.imread(img0).shape
    
    sfx = ''
    
    if factor is not None:
        sfx = '_{}'.format(factor)
        _minify(basedir, dir_name=dir_name, factors=[factor])
        factor = factor
    elif height is not None:
        factor = sh[0] / float(height)
        width = int(sh[1] / factor)
        _minify(basedir, dir_name=dir_name, resolutions=[[height, width]])
        sfx = '_{}x{}'.format(width, height)
    elif width is not None:
        factor = sh[1] / float(width)
        height = int(sh[0] / factor)
        _minify(basedir, dir_name=dir_name, resolutions=[[height, width]])
        sfx = '_{}x{}'.format(width, height)
    else:
        factor = 1
    
    imgdir = os.path.join(basedir, dir_name + sfx)
    if not os.path.exists(imgdir):
        print( imgdir, 'does not exist, returning' )
        return
    
    imgfiles = [os.path.join(imgdir, f) for i, f in enumerate(sorted(os.listdir(imgdir))) if check_fn(f, i)]
    
    return imgfiles, factor
        
        
def _load_data(basedir, factor=None, width=None, height=None, load_imgs=True, fg_mask=False, use_depth=False):

    check_colorimg_fn = lambda f, i: f.endswith('JPG') or f.endswith('jpg') or f.endswith('png')

    if fg_mask:
        check_maskimg_fn = lambda f, i: f.endswith('JPG') or f.endswith('jpg') or f.endswith('png')

    if use_depth:
        check_depthimg_fn = lambda f, i: f.endswith('JPG') or f.endswith('jpg') or f.endswith('png')
    
    poses_arr = np.load(os.path.join(basedir, 'poses_bounds.npy'))
    poses = poses_arr[:, :-2].reshape([-1, 3, 5]).transpose([1,2,0])
    bds = poses_arr[:, -2:].transpose([1,0])
    
    # img0 = [os.path.join(basedir, 'images', f) for f in sorted(os.listdir(os.path.join(basedir, 'images'))) if check_fn(f, 0)][0]
    # sh = imageio.imread(img0).shape
    
    # sfx = ''
    
    # if factor is not None:
    #     sfx = '_{}'.format(factor)
    #     _minify(basedir, factors=[factor])
    #     factor = factor
    # elif height is not None:
    #     factor = sh[0] / float(height)
    #     width = int(sh[1] / factor)
    #     _minify(basedir, resolutions=[[height, width]])
    #     sfx = '_{}x{}'.format(width, height)
    # elif width is not None:
    #     factor = sh[1] / float(width)
    #     height = int(sh[0] / factor)
    #     _minify(basedir, resolutions=[[height, width]])
    #     sfx = '_{}x{}'.format(width, height)
    # else:
    #     factor = 1
    
    # imgdir = os.path.join(basedir, 'images' + sfx)
    # if not os.path.exists(imgdir):
    #     print( imgdir, 'does not exist, returning' )
    #     return
    
    # imgfiles = [os.path.join(imgdir, f) for i, f in enumerate(sorted(os.listdir(imgdir))) if check_fn(f, i)]

    rgb_files, new_factor = _preprocess_imgs(basedir, dir_name='images', factor=factor, width=width, height=height, check_fn=check_colorimg_fn)

    if poses.shape[-1] != len(rgb_files):
        print( 'Mismatch between imgs {} and poses {} !!!!'.format(len(rgb_files), poses.shape[-1]))
        return
    
    sh = imageio.imread(rgb_files[0]).shape
    poses[:2, 4, :] = np.array(sh[:2]).reshape([2, 1])
    poses[2, 4, :] = poses[2, 4, :] * 1. / new_factor
    
    if not load_imgs:
        return poses, bds
    
    def imread(f):
        if f.endswith('png'):
            return imageio.imread(f, ignoregamma=True)
        else:
            return imageio.imread(f)
        
    rgb_imgs = [imread(f)[...,:3]/255. for f in rgb_files]
    rgb_imgs = np.stack(rgb_imgs, -1)

    mask_imgs = None
    if fg_mask:
        mask_files, _ = _preprocess_imgs(basedir, dir_name='masks', factor=factor, width=width, height=height, check_fn=check_maskimg_fn)
        
        if len(mask_files) != len(rgb_files):
            print( 'Mismatch between rgb imgs {} and mask imgs {} !!!!'.format(len(rgb_files), len(mask_files)))
            return
        
        mask_imgs = [imread(f) / 255.0 for f in mask_files]

        if mask_imgs[0].shape[:2] != rgb_imgs[..., 0].shape[:2]:
            print( 'Mismatch size between rgb imgs {} and mask imgs {} !!!!'.format(rgb_imgs[..., 0].shape[:2], mask_imgs[0].shape[:2]))
            return

        mask_imgs = np.stack(mask_imgs, -1)
        # Convert 0 for tool, 1 for not tool
        mask_imgs = 1.0 - mask_imgs

    depth_imgs = None
    if use_depth:
        depth_files, _ = _preprocess_imgs(basedir, dir_name='depth', factor=factor, width=width, height=height, check_fn=check_depthimg_fn)
        
        if len(depth_files) != len(rgb_files):
            print( 'Mismatch between rgb imgs {} and depth imgs {} !!!!'.format(len(rgb_files), len(depth_files)))
            return
        
        depth_imgs = [imread(f) for f in depth_files]

        if depth_imgs[0].shape[:2] != rgb_imgs[..., 0].shape[:2]:
            print( 'Mismatch size between rgb imgs {} and depth imgs {} !!!!'.format(rgb_imgs[..., 0].shape[:2], depth_imgs[0].shape[:2]))
            return
        
        depth_imgs = np.stack(depth_imgs, -1)
        # min_depth = np.percentile(depth_imgs, 3.0)
        # max_depth = np.percentile(depth_imgs, 99.9)
        # print('min depth:', min_depth, 'max depth:', max_depth)
    
    print('Loaded image data', rgb_imgs.shape, poses[:,-1,0])
    return poses, bds, rgb_imgs, mask_imgs, depth_imgs



def normalize(x):
    return x / np.linalg.norm(x)

def viewmatrix(z, up, pos):
    vec2 = normalize(z)
    vec1_avg = up
    vec0 = normalize(np.cross(vec1_avg, vec2))
    vec1 = normalize(np.cross(vec2, vec0))
    m = np.stack([vec0, vec1, vec2, pos], 1)
    return m

def ptstocam(pts, c2w):
    tt = np.matmul(c2w[:3,:3].T, (pts-c2w[:3,3])[...,np.newaxis])[...,0]
    return tt

def poses_avg(poses):

    hwf = poses[0, :3, -1:]

    center = poses[:, :3, 3].mean(0)
    vec2 = normalize(poses[:, :3, 2].sum(0))
    up = poses[:, :3, 1].sum(0)
    c2w = np.concatenate([viewmatrix(vec2, up, center), hwf], 1)
    
    return c2w



def render_path_spiral(c2w, up, rads, focal, zdelta, zrate, rots, N):
    render_poses = []
    rads = np.array(list(rads) + [1.])
    hwf = c2w[:,4:5]
    
    for theta in np.linspace(0., 2. * np.pi * rots, N+1)[:-1]:
        c = np.dot(c2w[:3,:4], np.array([np.cos(theta), -np.sin(theta), -np.sin(theta*zrate), 1.]) * rads) 
        z = normalize(c - np.dot(c2w[:3,:4], np.array([0,0,-focal, 1.])))
        render_poses.append(np.concatenate([viewmatrix(z, up, c), hwf], 1))
    return render_poses


def render_path_fixed(c2w, N):
    render_poses = []
    hwf = c2w[:,4:5]

    for i in range(N):
        eye_pose = np.array([[1.0, 0.0, 0.0, 0.0], [0.0, 1.0, 0.0, 0.0], [0.0, 0.0, 1.0, 0.0]])
        render_poses.append(np.concatenate([eye_pose, hwf], 1))

    return render_poses


def render_path_zoom(c2w, up, delta, N):
    render_poses = []
    hwf = c2w[:,4:5]
    z_d = c2w[:3,2]
    c_o = c2w[:3,3]

    half_N = (N+1) // 2
    
    for t in np.linspace(0., delta, half_N):
        c = c_o - t * z_d
        
        render_poses.append(np.concatenate([viewmatrix(z_d, up, c), hwf], 1))

    for t in np.linspace(0., delta, N+1 - half_N):
        c = c_o - (delta - t) * z_d
        
        render_poses.append(np.concatenate([viewmatrix(z_d, up, c), hwf], 1))

    return render_poses




def recenter_poses(poses):

    poses_ = poses+0
    bottom = np.reshape([0,0,0,1.], [1,4])
    c2w = poses_avg(poses)
    c2w = np.concatenate([c2w[:3,:4], bottom], -2)
    bottom = np.tile(np.reshape(bottom, [1,1,4]), [poses.shape[0],1,1])
    poses = np.concatenate([poses[:,:3,:4], bottom], -2)

    poses = np.linalg.inv(c2w) @ poses
    poses_[:,:3,:4] = poses[:,:3,:4]
    poses = poses_
    return poses


#####################


def spherify_poses(poses, bds):
    
    p34_to_44 = lambda p : np.concatenate([p, np.tile(np.reshape(np.eye(4)[-1,:], [1,1,4]), [p.shape[0], 1,1])], 1)
    
    rays_d = poses[:,:3,2:3]
    rays_o = poses[:,:3,3:4]

    def min_line_dist(rays_o, rays_d):
        A_i = np.eye(3) - rays_d * np.transpose(rays_d, [0,2,1])
        b_i = -A_i @ rays_o
        pt_mindist = np.squeeze(-np.linalg.inv((np.transpose(A_i, [0,2,1]) @ A_i).mean(0)) @ (b_i).mean(0))
        return pt_mindist

    pt_mindist = min_line_dist(rays_o, rays_d)
    
    center = pt_mindist
    up = (poses[:,:3,3] - center).mean(0)

    vec0 = normalize(up)
    vec1 = normalize(np.cross([.1,.2,.3], vec0))
    vec2 = normalize(np.cross(vec0, vec1))
    pos = center
    c2w = np.stack([vec1, vec2, vec0, pos], 1)

    poses_reset = np.linalg.inv(p34_to_44(c2w[None])) @ p34_to_44(poses[:,:3,:4])

    rad = np.sqrt(np.mean(np.sum(np.square(poses_reset[:,:3,3]), -1)))
    
    sc = 1./rad
    poses_reset[:,:3,3] *= sc
    bds *= sc
    rad *= sc
    
    centroid = np.mean(poses_reset[:,:3,3], 0)
    zh = centroid[2]
    radcircle = np.sqrt(rad**2-zh**2)
    new_poses = []
    
    for th in np.linspace(0.,2.*np.pi, 120):

        camorigin = np.array([radcircle * np.cos(th), radcircle * np.sin(th), zh])
        up = np.array([0,0,-1.])

        vec2 = normalize(camorigin)
        vec0 = normalize(np.cross(vec2, up))
        vec1 = normalize(np.cross(vec2, vec0))
        pos = camorigin
        p = np.stack([vec0, vec1, vec2, pos], 1)

        new_poses.append(p)

    new_poses = np.stack(new_poses, 0)
    
    new_poses = np.concatenate([new_poses, np.broadcast_to(poses[0,:3,-1:], new_poses[:,:3,-1:].shape)], -1)
    poses_reset = np.concatenate([poses_reset[:,:3,:4], np.broadcast_to(poses[0,:3,-1:], poses_reset[:,:3,-1:].shape)], -1)
    
    return poses_reset, new_poses, bds
    

def load_llff_data(basedir, factor=8, recenter=True, bd_factor=.75, spherify=False, path_zflat=False, davinci_endoscopic=False, fg_mask=False, render_path='spiral', use_depth=False):
    # No downsampling
    if factor == 1:
        factor = None

    if davinci_endoscopic:
        bd_factor = None


    poses, bds, imgs, masks, depth = _load_data(basedir, factor=factor, fg_mask=fg_mask, use_depth=use_depth) # factor=8 downsamples original imgs by 8x
    print('Loaded', basedir, bds.min(), bds.max())
    
    # Correct rotation matrix ordering and move variable dim to axis 0
    if not davinci_endoscopic:
        poses = np.concatenate([poses[:, 1:2, :], -poses[:, 0:1, :], poses[:, 2:, :]], 1)
    poses = np.moveaxis(poses, -1, 0).astype(np.float32)
    images = np.moveaxis(imgs, -1, 0).astype(np.float32)
    bds = np.moveaxis(bds, -1, 0).astype(np.float32)
    if fg_mask:
        masks = np.moveaxis(masks, -1, 0).astype(np.float32)
    if use_depth:
        depth = np.moveaxis(depth, -1, 0).astype(np.float32)
    
    # Rescale if bd_factor is provided
    sc = 1. if bd_factor is None else 1./(bds.min() * bd_factor + 1e-6)
    poses[:,:3,3] *= sc
    bds *= sc
    
    if recenter and not davinci_endoscopic:
        poses = recenter_poses(poses)
        
    if spherify:
        poses, render_poses, bds = spherify_poses(poses, bds)

    else:        
        c2w = poses_avg(poses)

        print('recentered', c2w.shape)
        print(c2w[:3,:4])

        ## Get spiral
        # Get average pose
        up = normalize(poses[:, :3, 1].sum(0))

        # Find a reasonable "focus depth" for this dataset
        close_depth, inf_depth = bds.min()*.9, bds.max()*5.
        dt = .75
        mean_dz = 1./(((1.-dt)/(close_depth + 1e-6) + dt/(inf_depth + 1e-6)) + 1e-6)
        focal = mean_dz

        # Get radii for spiral path
        shrink_factor = .8
        zdelta = close_depth * .2
        tt = poses[:,:3,3] # ptstocam(poses[:3,3,:].T, c2w).T
        rads = np.percentile(np.abs(tt), 90, 0)
        c2w_path = c2w
        N_views = 120
        N_rots = 2
        if path_zflat:
#             zloc = np.percentile(tt, 10, 0)[2]
            zloc = -close_depth * .1
            c2w_path[:3,3] = c2w_path[:3,3] + zloc * c2w_path[:3,2]
            rads[2] = 0.
            N_rots = 1
            N_views/=2

        if davinci_endoscopic:
            N_views = poses.shape[0]

        if render_path == 'spiral':
            if davinci_endoscopic:
                focal = 0.5
                rads = np.array([0.04, 0.05, 0])

            # Generate poses for spiral path
            render_poses = render_path_spiral(c2w_path, up, rads, focal, zdelta, zrate=.5, rots=N_rots, N=N_views)
            print('Render poses: spiral')
        elif render_path == 'fixidentity':
            # Generate poses for fixed path
            render_poses = render_path_fixed(c2w_path, N_views)
            print('Render poses: fix identity')
        elif render_path == 'zoom':
            zoom_dist = rads[2]
            if davinci_endoscopic:
                zoom_dist = 0.35

            # Generate poses for zoom path
            render_poses = render_path_zoom(c2w_path, up, zoom_dist, N_views)
            print('Render poses: zoom')
    
    render_poses = np.array(render_poses).astype(np.float32)

    c2w = poses_avg(poses)
    print('Data:', poses.shape, images.shape, bds.shape)
    
    dists = np.sum(np.square(c2w[:3,3] - poses[:,:3,3]), -1)
    i_test = np.argmin(dists)
    print('HOLDOUT view is', i_test)
    
    images = images.astype(np.float32)
    poses = poses.astype(np.float32)

    times = np.linspace(0., 1., poses.shape[0])
    render_times = torch.linspace(0., 1., render_poses.shape[0])

    return images, masks, depth, poses, times, bds, render_poses, render_times, i_test