TransMASK/data.py at main · VT-Collab/TransMASK · 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
import os
from pathlib import Path
import h5py
import numpy as np
import torch
from torch.utils.data import Dataset
from aeon.distances import dtw_pairwise_distance
from tqdm import tqdm
from utils import *
from torchvision import transforms as T

import matplotlib.pyplot as plt


class Data(Dataset):
    def __init__(self, hdf5_path, use_images=False):
        super().__init__()

        self.hdf5_path = str(hdf5_path)
        self.use_images = use_images

        with h5py.File(self.hdf5_path, 'r') as f:
            self.demo_keys = sorted([k for k in f.keys() if 'demo' in k], key=lambda x: int(x.split('_')[1]))
            self.demo_lens = [f[k]['actions'].shape[0] for k in self.demo_keys]
            self.num_demos = len(self.demo_keys)

            self.action_shape = f[self.demo_keys[0]]['actions'].shape
            if 'feature' in f[self.demo_keys[0]]['obs'].keys():
                self.feature_shape = f[self.demo_keys[0]]['obs']['feature'].shape
            else:
                self.feature_shape = [0, 0, 0]

        self.cumulative_demo_lens = np.cumsum(self.demo_lens)

    def __len__(self):
        return int(self.cumulative_demo_lens[-1])

    def _get_hdf5(self):
        if not hasattr(self, '_hdf5'):
            self._hdf5 = h5py.File(self.hdf5_path, 'r')
        return self._hdf5

    def __getitem__(self, index):
        demo_idx = np.searchsorted(self.cumulative_demo_lens, index, side='right')
        local_idx = index if demo_idx == 0 else index - self.cumulative_demo_lens[demo_idx - 1]

        f = self._get_hdf5()
        demo = f[self.demo_keys[demo_idx]]

        if not self.use_images:
            obs = demo['obs']['observation'][local_idx]
            goal = demo['obs']['desired_goal'][local_idx]
            obs = np.concatenate((obs, goal), axis=-1)
            actions = demo['actions'][local_idx]
            static = ego = features = None
        else:
            actions = demo['actions'][local_idx]
            obs = demo['obs']['observation'][local_idx, :actions.shape[-1]]
            static = demo['obs']['static_image'][local_idx]
            ego = demo['obs']['ee_image'][local_idx]
            features = np.concatenate((demo['obs']['static_seg'][local_idx],
                                       demo['obs']['ee_seg'][local_idx]),
                                       axis=-1)

        out = {'observation': torch.tensor(obs, dtype=torch.float32),
               'action': torch.tensor(actions, dtype=torch.float32)}

        if self.use_images:
            out.update({'static_image': torch.tensor(static, dtype=torch.float32).permute(2, 0, 1) / 255.0,
                        'ee_image': torch.tensor(ego, dtype=torch.float32).permute(2, 0, 1) / 255.0,
                        'feature': torch.tensor(features, dtype=torch.float32)})
        return out


class DataSequence(Dataset):
    def __init__(self, hdf5_path, obs_horizon=8, pred_horizon=4, use_images=False):
        super().__init__()

        self.hdf5_path = str(hdf5_path)
        self.obs_horizon = obs_horizon
        self.pred_horizon = pred_horizon
        self.use_images = use_images

        with h5py.File(self.hdf5_path, 'r') as f:
            self.demo_keys = sorted([k for k in f.keys() if 'demo' in k], key=lambda x: int(x.split('_')[1]))
            self.demo_lens = [f[k]['actions'].shape[0] for k in self.demo_keys]
            self.num_demos = len(self.demo_keys)

            self.action_shape = f[self.demo_keys[0]]['actions'].shape
            if 'feature' in f[self.demo_keys[0]]['obs'].keys():
                self.feature_shape = f[self.demo_keys[0]]['obs']['feature'].shape
            else:
                self.feature_shape = [0, 0, 0]

        self.cumulative_demo_lens = np.cumsum(self.demo_lens)

    def __len__(self):
        return int(self.cumulative_demo_lens[-1])

    def _get_hdf5(self):
        if not hasattr(self, '_hdf5'):
            self._hdf5 = h5py.File(self.hdf5_path, 'r')
        return self._hdf5

    def __getitem__(self, index):
        demo_idx = np.searchsorted(self.cumulative_demo_lens, index, side='right')
        local_idx = index if demo_idx == 0 else index - self.cumulative_demo_lens[demo_idx - 1]

        f = self._get_hdf5()
        demo = f[self.demo_keys[demo_idx]]
        demo_len = self.demo_lens[demo_idx]

        obs_end = min(local_idx + self.obs_horizon, demo_len)
        act_end = min(local_idx + self.pred_horizon, demo_len)

        if not self.use_images:
            obs = demo['obs']['observation'][local_idx:obs_end]
            goal = demo['obs']['desired_goal'][local_idx:obs_end]
            obs = np.concatenate((obs, goal), axis=-1)
            actions = demo['actions'][local_idx:act_end]
            static = ego = feat = None
        else:
            actions = demo['actions'][local_idx:act_end]
            obs = demo['obs']['observation'][local_idx:obs_end, :actions.shape[-1]]
            static = demo['obs']['static_image'][local_idx:obs_end]
            ego = demo['obs']['ee_image'][local_idx:obs_end]
            feat = np.concatenate((demo['obs']['static_seg'][local_idx:obs_end],
                                   demo['obs']['ee_seg'][local_idx:obs_end]),
                                   axis=-1)

        if obs.shape[0] < self.obs_horizon:
            pad = self.obs_horizon - obs.shape[0]
            obs = np.concatenate([obs, np.repeat(obs[-1:], pad, axis=0)])
            if self.use_images:
                static = np.concatenate([static, np.repeat(static[-1:], pad, axis=0)])
                ego = np.concatenate([ego, np.repeat(ego[-1:], pad, axis=0)])
                feat = np.concatenate([feat, np.repeat(feat[-1:], pad, axis=0)])

        if actions.shape[0] < self.pred_horizon:
            pad = self.pred_horizon - actions.shape[0]
            actions = np.concatenate([actions, np.repeat(actions[-1:], pad, axis=0)])

        out = {'observation': torch.tensor(obs, dtype=torch.float32),
               'action': torch.tensor(actions, dtype=torch.float32)}

        if self.use_images:
            out.update({'static_image': torch.tensor(static, dtype=torch.float32).permute(0, 3, 1, 2) / 255.0,
                        'ee_image': torch.tensor(ego, dtype=torch.float32).permute(0, 3, 1, 2) / 255.0,
                        'feature': torch.tensor(feat, dtype=torch.float32)})
        return out