fusion.py

#!/usr/bin/env python
import glob


import argparse
import os.path as osp
import logging
import time
import sys
sys.path.insert(0, osp.dirname(__file__) + '/..')
import numpy as np
import torch
import torch.nn as nn
import cv2
import torch.nn.functional as F
from plyfile import PlyData, PlyElement
from PIL import Image
import re
import torch
import os
from IPython import embed





def read_pfm(filename):
    file = open(filename, 'rb')
    color = None
    width = None
    height = None
    scale = None
    endian = None

    header = file.readline().decode('utf-8').rstrip()
    if header == 'PF':
        color = True
    elif header == 'Pf':
        color = False
    else:
        raise Exception('Not a PFM file.')

    dim_match = re.match(r'^(\d+)\s(\d+)\s$', file.readline().decode('utf-8'))
    if dim_match:
        width, height = map(int, dim_match.groups())
    else:
        raise Exception('Malformed PFM header.')

    scale = float(file.readline().rstrip())
    if scale < 0:  # little-endian
        endian = '<'
        scale = -scale
    else:
        endian = '>'  # big-endian

    data = np.fromfile(file, endian + 'f')
    shape = (height, width, 3) if color else (height, width)

    data = np.reshape(data, shape)
    data = np.flipud(data)
    file.close()
    return data, scale

def read_img(filename):
    img = Image.open(filename)
    np_img = np.array(img, dtype=np.float32) / 255.
    return np_img

def read_camera_parameters(filename):
    with open(filename) as f:
        lines = f.readlines()
        lines = [line.rstrip() for line in lines]
    # extrinsics: line [1,5), 4x4 matrix
    extrinsics = np.fromstring(' '.join(lines[1:5]), dtype=np.float32, sep=' ').reshape((4, 4))
    # intrinsics: line [7-10), 3x3 matrix
    intrinsics = np.fromstring(' '.join(lines[7:10]), dtype=np.float32, sep=' ').reshape((3, 3))
    return intrinsics, extrinsics

def read_pair_file(filename):
    data = []
    with open(filename) as f:
        num_viewpoint = int(f.readline())
        # 49 viewpoints
        # for view_idx in range(0,5):
        for view_idx in range(num_viewpoint):
            ref_view = int(f.readline().rstrip())
            src_views = [int(x) for x in f.readline().rstrip().split()[1::2]]
            data.append((ref_view, src_views))
    return data

def reproject_with_depth(depth_ref, intrinsics_ref, extrinsics_ref, depth_src, intrinsics_src, extrinsics_src):
    width, height = depth_ref.shape[1], depth_ref.shape[0]
    ## step1. project reference pixels to the source view
    # reference view x, y
    x_ref, y_ref = np.meshgrid(np.arange(0, width), np.arange(0, height))
    x_ref, y_ref = x_ref.reshape([-1]), y_ref.reshape([-1])
    # reference 3D space
    xyz_ref = np.matmul(np.linalg.inv(intrinsics_ref),
                        np.vstack((x_ref, y_ref, np.ones_like(x_ref))) * depth_ref.reshape([-1]))
    # source 3D space
    xyz_src = np.matmul(np.matmul(extrinsics_src, np.linalg.inv(extrinsics_ref)),
                        np.vstack((xyz_ref, np.ones_like(x_ref))))[:3]
    # source view x, y
    K_xyz_src = np.matmul(intrinsics_src, xyz_src)
    xy_src = K_xyz_src[:2] / K_xyz_src[2:3]

    ## step2. reproject the source view points with source view depth estimation
    # find the depth estimation of the source view
    x_src = xy_src[0].reshape([height, width]).astype(np.float32)
    y_src = xy_src[1].reshape([height, width]).astype(np.float32)
    sampled_depth_src = cv2.remap(depth_src, x_src, y_src, interpolation=cv2.INTER_LINEAR)
    # mask = sampled_depth_src > 0

    # source 3D space
    # NOTE that we should use sampled source-view depth_here to project back
    xyz_src = np.matmul(np.linalg.inv(intrinsics_src),
                        np.vstack((xy_src, np.ones_like(x_ref))) * sampled_depth_src.reshape([-1]))
    # reference 3D space
    xyz_reprojected = np.matmul(np.matmul(extrinsics_ref, np.linalg.inv(extrinsics_src)),
                                np.vstack((xyz_src, np.ones_like(x_ref))))[:3]
    # source view x, y, depth
    depth_reprojected = xyz_reprojected[2].reshape([height, width]).astype(np.float32)
    K_xyz_reprojected = np.matmul(intrinsics_ref, xyz_reprojected)
    xy_reprojected = K_xyz_reprojected[:2] / K_xyz_reprojected[2:3]
    x_reprojected = xy_reprojected[0].reshape([height, width]).astype(np.float32)
    y_reprojected = xy_reprojected[1].reshape([height, width]).astype(np.float32)

    return depth_reprojected, x_reprojected, y_reprojected, x_src, y_src

def check_geometric_consistency(depth_ref, intrinsics_ref, extrinsics_ref, depth_src, intrinsics_src, extrinsics_src):
    width, height = depth_ref.shape[1], depth_ref.shape[0]
    x_ref, y_ref = np.meshgrid(np.arange(0, width), np.arange(0, height))
    depth_reprojected, x2d_reprojected, y2d_reprojected, x2d_src, y2d_src = reproject_with_depth(depth_ref, intrinsics_ref, extrinsics_ref,
                                                     depth_src, intrinsics_src, extrinsics_src)
    # check |p_reproj-p_1| < 1
    dist = np.sqrt((x2d_reprojected - x_ref) ** 2 + (y2d_reprojected - y_ref) ** 2)

    # check |d_reproj-d_1| / d_1 < 0.01
    depth_diff = np.abs(depth_reprojected - depth_ref)
    relative_depth_diff = depth_diff / depth_ref

    mask = np.logical_and(dist < 0.3, relative_depth_diff < 0.001)
    # mask = np.logical_and(dist < 1, relative_depth_diff < 0.01)
    depth_reprojected[~mask] = 0

    return mask, depth_reprojected, x2d_src, y2d_src

def filter_depth(scan_folder, pair_folder, plyfilename):
    # the pair file
    pair_file = os.path.join(pair_folder, "pair.txt")
    # for the final point cloud
    vertexs = []
    vertex_colors = []

    pair_data = read_pair_file(pair_file)
    nviews = len(pair_data)
    # TODO: hardcode size

    # for each reference view and the corresponding source views
    for ref_view, src_views in pair_data:
        # load the camera parameters
        ref_intrinsics, ref_extrinsics = read_camera_parameters(
            os.path.join(scan_folder, 'cam/cam_{:0>8}.txt'.format(ref_view)))
        # load the reference image
        ref_img = read_img(os.path.join(scan_folder, 'rgb/{:0>8}.png'.format(ref_view)))
        # load the estimated depth of the reference view
        ref_depth_est = read_pfm(os.path.join(scan_folder, 'depth_sdf/dep_{:0>8}.pfm'.format(ref_view)))[0]
        # load the photometric mask of the reference view
        confidence = read_pfm(os.path.join(scan_folder, 'confidence/conf_{:0>8}.pfm'.format(ref_view)))[0]
        h, w = ref_depth_est.shape

        confidence = cv2.resize(confidence, (w, h), interpolation=cv2.INTER_LINEAR)       
        photo_mask = confidence > 0.8

        all_srcview_depth_ests = []
        all_srcview_x = []
        all_srcview_y = []
        all_srcview_geomask = []

        # compute the geometric mask
        geo_mask_sum = 0
        for src_view in src_views:
            # camera parameters of the source view
            src_intrinsics, src_extrinsics = read_camera_parameters(
                os.path.join(scan_folder, 'cam/cam_{:0>8}.txt'.format(src_view)))
            # the estimated depth of the source view
            src_depth_est = read_pfm(os.path.join(scan_folder, 'depth_sdf/dep_{:0>8}.pfm'.format(src_view)))[0]

            geo_mask, depth_reprojected, x2d_src, y2d_src = check_geometric_consistency(ref_depth_est, ref_intrinsics, ref_extrinsics,
                                                                      src_depth_est,
                                                                      src_intrinsics, src_extrinsics)
            geo_mask_sum += geo_mask.astype(np.int32)
            all_srcview_depth_ests.append(depth_reprojected)
            all_srcview_x.append(x2d_src)
            all_srcview_y.append(y2d_src)
            all_srcview_geomask.append(geo_mask)

        depth_est_averaged = (sum(all_srcview_depth_ests) + ref_depth_est) / (geo_mask_sum + 1)
        # at least 3 source views matched
        geo_mask = geo_mask_sum >= 3
        final_mask = np.logical_and(photo_mask, geo_mask)

        print("processing {}, ref-view{:0>2}, photo/geo/final-mask:{}/{}/{}".format(scan_folder, ref_view,
                                                                                    photo_mask.mean(),
                                                                                    geo_mask.mean(), final_mask.mean()))
        height, width = depth_est_averaged.shape[:2]
        x, y = np.meshgrid(np.arange(0, width), np.arange(0, height))
        valid_points = final_mask
        print("valid_points", valid_points.mean())
        x, y, depth = x[valid_points], y[valid_points], depth_est_averaged[valid_points]

        color = ref_img[valid_points]  # hardcoded for DTU dataset

        xyz_ref = np.matmul(np.linalg.inv(ref_intrinsics),
                            np.vstack((x, y, np.ones_like(x))) * depth)
        xyz_world = np.matmul(np.linalg.inv(ref_extrinsics),
                              np.vstack((xyz_ref, np.ones_like(x))))[:3]
        vertexs.append(xyz_world.transpose((1, 0)))
        vertex_colors.append((color * 255).astype(np.uint8))

    vertexs = np.concatenate(vertexs, axis=0)
    vertex_colors = np.concatenate(vertex_colors, axis=0)
    vertexs = np.array([tuple(v) for v in vertexs], dtype=[('x', 'f4'), ('y', 'f4'), ('z', 'f4')])
    vertex_colors = np.array([tuple(v) for v in vertex_colors], dtype=[('red', 'u1'), ('green', 'u1'), ('blue', 'u1')])

    vertex_all = np.empty(len(vertexs), vertexs.dtype.descr + vertex_colors.dtype.descr)
    for prop in vertexs.dtype.names:
        vertex_all[prop] = vertexs[prop]
    for prop in vertex_colors.dtype.names:
        vertex_all[prop] = vertex_colors[prop]

    el = PlyElement.describe(vertex_all, 'vertex')
    PlyData([el]).write(plyfilename)
    print("saving the final model to", plyfilename)

if __name__ == "__main__":

    root_dir = './outputs'

    testlist='./dataloader/datalist/dtu/test.txt'

    testpath='./data/dtu/Eval'

    with open(testlist) as f:
        scans = f.readlines()
        scans = [line.rstrip() for line in scans]

    for scan in scans:
        scan_id = int(scan[4:])
        scan_folder = os.path.join(root_dir, scan)
        pair_folder = os.path.join(testpath, scan)

        filter_depth(scan_folder, pair_folder, os.path.join(root_dir, 'sdfnet{:0>3}_l3.ply'.format(scan_id)))