File size: 6,354 Bytes
f0d013b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
""" GraspNet dataset processing.
"""

import os
import sys
import numpy as np
import numpy.ma as ma
import scipy.io as scio
from scipy.optimize import linear_sum_assignment
from PIL import Image
from skimage.measure import label, regionprops
import cv2

import torch
from collections import abc as container_abcs
from torch.utils.data import Dataset
from tqdm import tqdm
from torch.utils.data import DataLoader
from time import time

BASE_DIR = os.path.dirname(os.path.abspath(__file__))
from .data_utils import CameraInfo, transform_point_cloud, create_point_cloud_from_depth_image, \
    get_workspace_mask, remove_invisible_grasp_points
import h5py


class GraspNetDataset(Dataset):
    def __init__(self, root, valid_obj_idxs, camera='kinect', split='train', remove_invisible=True,
                 augment=False, limited_data=False, overfitting=False, k_grasps=1, ground_truth_type="topk", caching=True):
        self.root = root
        self.split = split
        self.remove_invisible = remove_invisible
        self.valid_obj_idxs = valid_obj_idxs
        self.camera = camera
        self.augment = augment
        self.k_grasps = k_grasps
        self.ground_truth_type = ground_truth_type
        self.overfitting = overfitting
        self.caching = caching

        if overfitting:
            limited_data = True
        self.limited_data = limited_data

        if split == 'train':
            self.sceneIds = list(range(100))
        elif split == 'test':
            self.sceneIds = list(range(100, 190))
        elif split == 'test_seen':
            self.sceneIds = list(range(100, 130))
        elif split == 'test_similar':
            self.sceneIds = list(range(130, 160))
        elif split == 'test_novel':
            self.sceneIds = list(range(160, 190))
        if limited_data:
            self.sceneIds = self.sceneIds[:10]
        self.sceneIds = ['scene_{}'.format(str(x).zfill(4)) for x in self.sceneIds]

        filename = f"dataset/{split}_labels"
        if limited_data and not overfitting:
            filename += "_limited"
        if overfitting:
            filename += "_overfitting"
        filename += ".hdf5"
        self.h5_filename = filename
        self.h5_file = None
        self.grasp_labels_filename = "dataset/grasp_labels.hdf5"
        self.grasp_labels_file = None

        with h5py.File(self.h5_filename, 'r') as f:
            self.len = f['depthpath'].shape[0]

    def __len__(self):
        return self.len

    def __getitem__(self, index):
        if self.h5_file is None:
            self.h5_file = h5py.File(self.h5_filename, 'r')

        ann_id = int(str(self.h5_file['metapath'][index], 'utf-8').split("meta")[1][1:-4])

        color = np.array(Image.open(self.h5_file['colorpath'][index]), dtype=np.float32) / 255.0
        depth = np.array(Image.open(self.h5_file['depthpath'][index]))

        # fixing depth image where value is 0
        p99 = np.percentile(depth[depth != 0], 99)
        # p1 = abs(np.percentile(depth[depth != 0], 1))
        depth[depth > p99] = p99
        depth[depth == 0] = p99

        seg = np.array(Image.open(self.h5_file['labelpath'][index]))
        meta = scio.loadmat(self.h5_file['metapath'][index])
        scene = self.h5_file['scenename'][index]

        main_path = str(self.h5_file['metapath'][index], 'utf-8').split("meta")[0]
        cam_extrinsics = np.load(os.path.join(str(self.h5_file['metapath'][index], 'utf-8').split("meta")[0],
                                              'camera_poses.npy'))[ann_id]
        cam_wrt_table = np.load(os.path.join(str(self.h5_file['metapath'][index], 'utf-8').split("meta")[0],
                                             'cam0_wrt_table.npy'))
        cam_extrinsics = cam_wrt_table.dot(cam_extrinsics).astype(np.float32)

        try:
            obj_idxs = meta['cls_indexes'].flatten().astype(np.int32)
            poses = meta['poses']
            intrinsic = meta['intrinsic_matrix']
            factor_depth = meta['factor_depth']
        except Exception as e:
            print(repr(e))
            print(scene)

        # h_ratio = 800 / 720
        # w_ratio = 1333 / 1280

        camera = CameraInfo(1280.0, 720.0, intrinsic[0][0], intrinsic[1][1], intrinsic[0][2], intrinsic[1][2], factor_depth)

        ## generate cloud required to remove invisible grasp points
        #cloud = create_point_cloud_from_depth_image(depth, camera, organized=True)

        obj_bounding_boxes = []
        for i, obj_idx in enumerate(obj_idxs):
            if obj_idx not in self.valid_obj_idxs:
                continue
            if (seg == obj_idx).sum() < 50:
                continue

            seg_cpy = seg.copy()
            seg_cpy[seg != obj_idx] = 0
            seg_cpy[seg == obj_idx] = 1
            seg_labels = label(seg_cpy)
            regions = regionprops(seg_labels)

            # b has start_height, start_width, end_height, end_width = (x_min, y_min, x_max, y_max)
            b = regions[0].bbox
            # saved bbox has xyxy
            H, W = seg.shape[0], seg.shape[1]

            obj_bounding_boxes.append(np.array([b[1] / W, b[0] / H, b[3] / W, b[2] / H])[None].repeat(self.k_grasps, 0))
        obj_bounding_boxes = np.concatenate(obj_bounding_boxes, axis=0).astype(np.float32)

        ret_dict = {}
        #ret_dict['point_cloud'] = cloud.transpose((2, 0, 1)).astype(np.float32)
        ret_dict['color'] = color.transpose((2, 0, 1)).astype(np.float32)
        ret_dict['depth'] = (depth / camera.scale).astype(np.float32)
        ret_dict['objectness_label'] = seg.astype(np.int32)
        ret_dict['obj_bounding_boxes'] = obj_bounding_boxes
        ret_dict['camera_intrinsics'] = np.expand_dims(np.concatenate([intrinsic.reshape(-1), factor_depth[0]]), -1).astype(np.float32)
        ret_dict['camera_extrinsics'] = cam_extrinsics.astype(np.float32)
        #ret_dict['transformed_points'] = transformed_points.astype(np.float32)
        ret_dict['obj_idxs'] = obj_idxs

        return ret_dict


def load_valid_obj_idxs():
    obj_names = list(range(88))
    valid_obj_idxs = []
    for i, obj_name in enumerate(obj_names):
        if i == 18: continue
        valid_obj_idxs.append(i + 1)  # here align with label png

    return valid_obj_idxs


def my_worker_init_fn(worker_id):
    np.random.seed(np.random.get_state()[1][0] + worker_id)
    pass