-
Notifications
You must be signed in to change notification settings - Fork 18
/
Copy pathload_dataset.py
131 lines (100 loc) · 4.96 KB
/
load_dataset.py
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
from functools import reduce
import numpy as np
from general_utils import get_data_dir
from general_utils import get_image_size
from grid import grid_mask
from load_model_config import ModelConfig
from preprocessors.preprocessors_utils import create_dataset_preprocessor
def load_dataset_from_config(config: ModelConfig):
data_dir = get_data_dir(config.data_root, config.test_dataset_kind)
return load_dataset(data_dir=data_dir,
dataset_kind=config.test_dataset_kind,
seq_len=config.obs_len + config.pred_len,
max_n_peds=config.max_n_peds,
n_neighbor_pixels=config.n_neighbor_pixels,
grid_side=config.grid_side)
def load_dataset(data_dir, dataset_kind, seq_len, max_n_peds, n_neighbor_pixels,
grid_side):
loader = _SingleDatasetLoader(data_dir, seq_len, max_n_peds,
n_neighbor_pixels, grid_side,
dataset_kind)
dataset = loader.load()
return dataset
class SingleDataset:
def __init__(self, frame_data, seq_len, max_n_peds, n_neighbor_pixels,
grid_side, image_size):
self.seq_len = seq_len
self.max_n_peds = max_n_peds
self.n_neighbor_pixels = n_neighbor_pixels
self.grid_side = grid_side
self.image_size = image_size
self.x_data, self.y_data, self.grid_data = self._build_data(frame_data)
def _build_data(self, frame_data):
x_data = []
y_data = []
for i in range(len(frame_data) - self.seq_len):
cf_data = frame_data[i:i + self.seq_len, ...]
nf_data = frame_data[i + 1:i + self.seq_len + 1, ...]
ped_col_index = 0
# collect ped ids where the ped id exists in the all frame of
# the current sequence and the next sequence
cf_ped_ids = reduce(set.intersection,
[set(nf_ped_ids) for nf_ped_ids in
cf_data[..., ped_col_index]])
nf_ped_ids = reduce(set.intersection,
[set(nf_ped_ids) for nf_ped_ids in
nf_data[..., ped_col_index]])
ped_ids = list(cf_ped_ids & nf_ped_ids - {0})
# at current & next frame, there are no pedestrians
if not ped_ids:
continue
x = np.zeros((self.seq_len, self.max_n_peds, 3))
y = np.zeros((self.seq_len, self.max_n_peds, 3))
# fi = frame index, cf = current frame, nf = next frame
for fi, (cf, nf) in enumerate(zip(cf_data, nf_data)):
for j, ped_id in enumerate(ped_ids):
cf_ped_row = cf[:, 0] == ped_id
nf_ped_row = nf[:, 0] == ped_id
if np.any(cf_ped_row):
x[fi, j, :] = cf[cf[:, 0] == ped_id]
if np.any(nf_ped_row):
y[fi, j, :] = nf[nf[:, 0] == ped_id]
x_data.append(x)
y_data.append(y)
# compute grid mask
grid_data = [grid_mask(x, self.image_size, self.n_neighbor_pixels,
self.grid_side) for x in x_data]
data_tuple = (np.array(x_data, np.float32),
np.array(y_data, np.float32),
np.array(grid_data, np.float32))
return data_tuple
def get_data(self, lstm_state_dim):
zeros_data = np.zeros((len(self.x_data), self.seq_len,
self.max_n_peds, lstm_state_dim), np.float32)
return self.x_data, self.y_data, self.grid_data, zeros_data
class _SingleDatasetLoader:
def __init__(self, data_dir, seq_len, max_n_peds, n_neighbor_pixels,
grid_side, dataset_kind):
self.data_dir = data_dir
self.seq_len = seq_len
self.max_n_peds = max_n_peds
self.n_neighbor_pixels = n_neighbor_pixels
self.grid_side = grid_side
self.dataset_kind = dataset_kind
self.image_size = get_image_size(dataset_kind)
def load(self) -> SingleDataset:
preprocessor = create_dataset_preprocessor(self.data_dir,
self.dataset_kind)
df = preprocessor.preprocess_frame_data()
# All frame IDs in the current dataset
all_frames = df["frame"].unique().tolist()
n_all_frames = len(all_frames)
all_frame_data = np.zeros((n_all_frames, self.max_n_peds, 3),
np.float64)
for index, frame in enumerate(all_frames):
peds_with_pos = np.array(df[df["frame"] == frame][["id", "x", "y"]])
n_peds = len(peds_with_pos)
all_frame_data[index, 0:n_peds, :] = peds_with_pos
return SingleDataset(all_frame_data, self.seq_len,
self.max_n_peds, self.n_neighbor_pixels,
self.grid_side, self.image_size)