SimpleNeRF/src/data_loaders/RealEstateDataLoader01.py at main · NagabhushanSN95/SimpleNeRF · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
# Shree KRISHNAYa Namaha
# Loads RealEstate Data for NeRF, Colmap sparse depth, dense depth, visibility prior from RealEstate-10K dataset.
# Authors: Nagabhushan S N, Adithyan K V
# Last Modified: 15/06/2023

from pathlib import Path
from typing import Optional

import numpy
import pandas
import skimage.io

from data_loaders.DataLoaderParent import DataLoaderParent


class RealEstateDataLoader(DataLoaderParent):
    def __init__(self, configs: dict, data_dirpath: Path, mode: Optional[str]):
        super(RealEstateDataLoader, self).__init__()
        self.configs = configs
        self.data_dirpath = data_dirpath
        self.mode = mode
        self.scene_num = int(configs['data_loader']['scene_id'])

        self.sparse_depth_needed = 'sparse_depth' in self.configs['data_loader']
        self.dense_depth_needed = ('dense_depth' in self.configs['data_loader'])
        self.visibility_prior_needed = ('visibility_prior' in self.configs['data_loader'])
        return

    def load_data(self):
        frame_nums = self.get_frame_nums()
        data_dict = {
            'frame_nums': frame_nums,
        }

        data_dict['nerf_data'] = self.load_nerf_data(data_dict)

        if self.sparse_depth_needed and self.mode == 'train':
            data_dict['sparse_depth_data'] = self.load_sparse_depth_data(data_dict)

        if self.dense_depth_needed and self.mode == 'train':
            data_dict['dense_depth_data'] = self.load_dense_depth_data(data_dict)

        if self.visibility_prior_needed and self.mode == 'train':
            data_dict['visibility_prior_data'] = self.load_visibility_prior_data(data_dict)

        return data_dict

    def get_frame_nums(self):
        set_num = self.configs['data_loader']['train_set_num']
        video_datapath = self.data_dirpath / f'train_test_sets/set{set_num:02}/{self.mode.capitalize()}VideosData.csv'
        video_data = pandas.read_csv(video_datapath)
        frame_nums = video_data.loc[video_data['scene_num'] == self.scene_num]['pred_frame_num'].to_numpy()
        return frame_nums

    def load_nerf_data(self, data_dict):
        frame_nums = data_dict['frame_nums']
        images_dirpath = self.data_dirpath / f'test/database_data/{self.scene_num:05}/rgb'
        if not images_dirpath.exists():
            print(f'{images_dirpath.as_posix()} does not exist, returning.')
            return
        images_paths = [images_dirpath / f'{frame_num:04}.png' for frame_num in frame_nums]
        images = [self.read_image(image_path) for image_path in images_paths]
        images = numpy.stack(images)

        bounds = numpy.array([1, 100]).astype('float32')

        extrinsics_path = self.data_dirpath / f'test/database_data/{self.scene_num:05}/CameraExtrinsics.csv'
        extrinsic_matrices = numpy.loadtxt(extrinsics_path.as_posix(), delimiter=',').reshape((-1, 4, 4))
        extrinsics = extrinsic_matrices[frame_nums]

        intrinsics_path = self.data_dirpath / f'test/database_data/{self.scene_num:05}/CameraIntrinsics.csv'
        intrinsic_matrices = numpy.loadtxt(intrinsics_path.as_posix(), delimiter=',').reshape((-1, 3, 3))
        intrinsics = intrinsic_matrices[frame_nums]

        h, w = images.shape[1:3]
        return_dict = {
            'images': images,
            'extrinsics': extrinsics,
            'intrinsics': intrinsics,
            'resolution': (h, w),
            'bounds': bounds,
        }
        return return_dict

    def load_sparse_depth_data(self, data_dict: dict):
        sparse_depth_data = {}
        depth_dirname = self.configs['data_loader']['sparse_depth']['dirname']
        for frame_num in data_dict['frame_nums']:
            depth_path = self.data_dirpath / f'test/estimated_depths/{depth_dirname}/{self.scene_num:05}/estimated_depths/{frame_num:04}.csv'
            depth_data = pandas.read_csv(depth_path)
            sparse_depth_data[frame_num] = depth_data
        return sparse_depth_data

    def load_dense_depth_data(self, data_dict: dict):
        dense_depth_data = {}

        h, w = data_dict['nerf_data']['resolution']
        depths, depth_weights = [], []
        depth_dirname = self.configs['data_loader']['dense_depth']['dirname']
        weights_suffix = ''
        if 'weights_suffix' in self.configs['data_loader']['dense_depth']:
            weights_suffix = self.configs['data_loader']['dense_depth']['weights_suffix']
        for frame_num in data_dict['frame_nums']:
            depth_path = self.data_dirpath / f'test/estimated_depths/{depth_dirname}/{self.scene_num:05}/estimated_depths/{frame_num:04}.npy'
            print(f'Loading depth: {depth_path.as_posix()}')

            depth = numpy.load(depth_path.as_posix())
            depths.append(depth)

            weights_path = self.data_dirpath / f'test/estimated_depths/{depth_dirname}/{self.scene_num:05}/Weights{weights_suffix}/{frame_num:04}.npy'
            if weights_path.exists():
                depth_weight = numpy.load(weights_path.as_posix())[:, :]
            else:
                print(f'Dense Depth Weights {weights_path.as_posix()} not found!. Loading unit weights.')
                depth_weight = numpy.ones(shape=(h, w))
            depth_weights.append(depth_weight)
        depths = numpy.stack(depths, axis=0)
        depth_weights = numpy.stack(depth_weights, axis=0)

        dense_depth_data['depth_values'] = depths
        dense_depth_data['depth_weights'] = depth_weights

        return dense_depth_data

    def load_visibility_prior_data(self, data_dict):
        visibility_prior_data = {}

        if self.configs['data_loader']['visibility_prior']['load_masks']:
            masks = []
            masks_dirname = self.configs['data_loader']['visibility_prior']['masks_dirname']
            frame1_nums = data_dict['frame_nums']
            for frame1_num in frame1_nums:
                frame2_nums = [x for x in frame1_nums if x != frame1_num]
                frame1_masks = []
                for frame2_num in frame2_nums:
                    mask_path = self.data_dirpath / f'test/visibility_prior/{masks_dirname}/{self.scene_num:05}/visibility_masks/{frame1_num:04}_{frame2_num:04}.png'
                    print(f'Loading visibility prior mask: {mask_path.as_posix()}')
                    mask = self.read_mask(mask_path)
                    frame1_masks.append(mask)
                masks.append(frame1_masks)
            masks = numpy.array(masks)  # (n, n-1, h, w)
            visibility_prior_data['masks'] = masks

        if self.configs['data_loader']['visibility_prior']['load_weights']:
            weights = []
            weights_dirname = self.configs['data_loader']['visibility_prior']['weights_dirname']
            frame1_nums = data_dict['frame_nums']
            for frame1_num in frame1_nums:
                frame2_nums = [x for x in frame1_nums if x != frame1_num]
                frame1_weights = []
                for frame2_num in frame2_nums:
                    weight_path = self.data_dirpath / f'test/visibility_prior/{weights_dirname}/{self.scene_num:05}/visibility_weights/{frame1_num:04}_{frame2_num:04}.npy'
                    print(f'Loading visibility prior weight: {weight_path.as_posix()}')
                    weight = self.read_npy_file(weight_path)
                    frame1_weights.append(weight)
                weights.append(frame1_weights)
            weights = numpy.array(weights)  # (n, n-1, h, w)
            visibility_prior_data['weights'] = weights
        return visibility_prior_data

    @staticmethod
    def read_image(path: Path, mmap_mode: str = None):
        if path.suffix in ['.png']:
            image = skimage.io.imread(path.as_posix())
        elif path.suffix == '.npy':
            image = numpy.load(path.as_posix(), mmap_mode=mmap_mode)
        else:
            raise RuntimeError(f'Unknown image format: {path.as_posix()}')
        return image

    @staticmethod
    def read_mask(path: Path, mmap_mode: str = None):
        if path.suffix in ['.png']:
            mask = skimage.io.imread(path.as_posix()) == 255
        elif path.suffix == '.npy':
            mask = numpy.load(path.as_posix(), mmap_mode=mmap_mode)
        else:
            raise RuntimeError(f'Unknown mask format: {path.as_posix()}')
        return mask

    @staticmethod
    def read_npy_file(path: Path, mmap_mode: str = None):
        if path.suffix == '.npy':
            data = numpy.load(path.as_posix(), mmap_mode=mmap_mode)
        else:
            raise RuntimeError(f'Unknown data format: {path.as_posix()}')
        return data