S3Net/data.py at main · CVEO/S3Net · 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
import os
import cv2
import torch
import random
import torchvision.transforms as transforms
import torch.distributed as dist
import numpy as np
from torch.utils.data import Dataset

class DataLoader:
    @staticmethod
    def load(datapath: str, dataset: str) -> tuple:
        def _get_image_paths(subdir):
            base_path = os.path.join(datapath, subdir)
            return [os.path.join(base_path, img) for img in os.listdir(base_path)]

        left_train = _get_image_paths("left")
        right_train = [path.replace("LEFT_RGB", "RIGHT_RGB").replace("left", "right") for path in left_train]
        disp_train = [path.replace("LEFT_RGB", "LEFT_DSP").replace("left", "disp") for path in left_train]
        cls_train = [path.replace("LEFT_RGB", "LEFT_CLS").replace("left", "cls") for path in left_train]

        left_valid = _get_image_paths("valid_left")
        right_valid = [path.replace("LEFT_RGB", "RIGHT_RGB").replace("left", "right") for path in left_valid]
        disp_valid = [path.replace("LEFT_RGB", "LEFT_DSP").replace("left", "disp") for path in left_valid]
        cls_valid = [path.replace("LEFT_RGB", "LEFT_CLS").replace("left", "cls") for path in left_valid]


        train_data = (left_train, right_train, disp_train, cls_train)
        valid_data = (left_valid, right_valid, disp_valid, cls_valid)

        return train_data, valid_data

class StereoDataset(Dataset):
    def __init__(self, left_images, right_images, disp_images, cls_images, training=True):
        self.left = left_images
        self.right = right_images
        self.disp = disp_images
        self.cls = cls_images
        self.training = training

    def __len__(self):
        return len(self.left)

    def get_transform(self, data):
        normal_mean_var = {'mean': [0.485, 0.456, 0.406],
                            'std': [0.229, 0.224, 0.225]}
        data = torch.from_numpy(data).float()
        transform = transforms.Compose([transforms.Normalize(**normal_mean_var)])
        return transform(data).float()

    def _augment(self, left, right, disp, cls):

        if random.random() > 0.5:
            left = np.flip(left, axis=1).copy()
            right = np.flip(right, axis=1).copy()
            disp = np.flip(disp, axis=0).copy()
            cls = np.flip(cls, axis=0).copy()

        if random.random() > 0.5:
            left = np.flip(left, axis=2).copy()
            right = np.flip(right, axis=2).copy()
            disp = -np.flip(disp, axis=1).copy()
            cls = np.flip(cls, axis=1).copy()

        _, h, w = left.shape
        x = random.randint(0, w - 512)
        y = random.randint(0, h - 512)
        left = left[:, y:y+512, x:x+512].copy()
        right = right[:, y:y+512, x:x+512].copy()
        cls = cls[y:y+512, x:x+512].copy()
        disp = disp[y:y+512, x:x+512].copy()

        return left, right, disp, cls

    def __getitem__(self, index):
        left = self._read_image(self.left[index])
        right = self._read_image(self.right[index])
        disp = self._read_image(self.disp[index], is_disp_cls=True)
        cls = self._read_image(self.cls[index], is_disp_cls=True)

        if self.training:
            left, right, disp, cls = self._augment(left, right, disp, cls)

        left = self.get_transform(left)
        right = self.get_transform(right)

        return left, right, disp, cls

    def _read_image(self, path, is_disp_cls=False):
        img = cv2.imread(path, cv2.IMREAD_UNCHANGED).astype('float32')

        if len(img.shape) == 3:
            img = np.moveaxis(img, -1, 0) / 255.0
            return img

        if is_disp_cls:
            return img


def generate(dataset, datapath):

    train_data, valid_data = DataLoader.load(datapath, dataset)

    train_dataset = StereoDataset(
        left_images=train_data[0],
        right_images=train_data[1],
        disp_images=train_data[2],
        cls_images=train_data[3],
        training=True
    )

    valid_dataset = StereoDataset(
        left_images=valid_data[0],
        right_images=valid_data[1],
        disp_images=valid_data[2],
        cls_images=valid_data[3],
        training=False
    )

    return train_dataset, valid_dataset


def initialize_dataloaders(args, train_dataset, valid_dataset):
    if args.is_distributed:
        train_sampler = torch.utils.data.DistributedSampler(train_dataset, num_replicas=dist.get_world_size(),
                                                            rank=dist.get_rank())
        valid_sampler = torch.utils.data.DistributedSampler(valid_dataset, num_replicas=dist.get_world_size(),
                                                            rank=dist.get_rank())
        train_loader = torch.utils.data.DataLoader(
            train_dataset, batch_size=args.batch_size, num_workers=args.num_workers,
            sampler=train_sampler, pin_memory=True)

        valid_loader = torch.utils.data.DataLoader(
            valid_dataset, batch_size=args.batch_size, num_workers=args.num_workers,
            sampler=valid_sampler, pin_memory=True)
    else:
        train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=args.batch_size,
                                                   shuffle=True, num_workers=args.num_workers, drop_last=False)
        valid_loader = torch.utils.data.DataLoader(valid_dataset, batch_size=args.batch_size,
                                                   shuffle=False, num_workers=args.num_workers, drop_last=False)
    return train_loader, valid_loader