image-driven/data.py at main · scottworkman/image-driven · 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
# Copyright © Scott Workman. 2024.

import torch
from torch.utils.data import Dataset
import torchvision.transforms.functional as TF

import utils

import os
import sparse
import numpy as np
import pandas as pd
from pathlib import Path


class DTSDataset(Dataset):
  """Dynamic Traffic Speeds Dataset."""

  def __init__(self,
               mode='train',
               dense=False,
               full=False,
               city="new_york",
               data_dir=None):
    if data_dir is None:
      data_dir = f"{Path(os.path.abspath(__file__)).parent}/dts++/"
    self.data_dir = data_dir
    self.mode = mode
    self.dense = dense
    self.full = full
    self.city = city
    self.df = self.__load_df()

  def __load_df(self):
    if self.mode == "train":
      names = ['image', 'label_mask', 'label_speed']
    else:
      names = ['image', 'label_mask', 'label_speed', 'dow', 'hour']

    df = pd.read_csv("{}{}_{}.txt".format(self.data_dir, self.mode, self.city),
                     sep=',',
                     header=None,
                     names=names)
    df['tile'] = df.apply(
        lambda row: [int(x) for x in row['image'][:-4].strip().split('/')[-3:]],
        axis=1)
    absolute = lambda x: os.path.join(self.data_dir, str(x))
    df['image'] = df['image'].apply(absolute)
    df['label_mask'] = df['label_mask'].apply(absolute)
    df['label_speed'] = df['label_speed'].apply(absolute)
    return df

  def __getitem__(self, idx):
    image = utils.preprocess(utils.imread(self.df['image'][idx]))
    mask = sparse.load_npz(self.df['label_mask'][idx]).todense()
    df_speed = pd.read_csv(self.df['label_speed'][idx])

    # extract labels from mask
    sliced = [x.squeeze() for x in np.array_split(mask, 6, axis=2)]
    l_road, l_road_id, _, l_samp_id, l_angle, l_angle_bin = sliced
    l_angle_bin = l_angle_bin - 1  # invalid pixels are now -1, 16 bins

    # sample a time
    if self.mode == 'train':
      time = utils.sample_time(df_speed)
    else:
      time = np.asarray([self.df['dow'][idx], self.df['hour'][idx]])

    # parameterize location
    location = utils.compute_location(self.df['tile'][idx],
                                      out_shape=image.shape,
                                      city=self.city)

    # generate speed mask
    result = df_speed.query('dow == {} and hour == {}'.format(time[0], time[1]))
    if self.dense:
      l_speed, _ = utils.generate_speed_mask(result['id'],
                                             result['speed_mph_mean'],
                                             l_road_id)
      if self.full:
        # don't modify the road ids
        l_count, _ = utils.generate_speed_mask(result['id'], result['count'],
                                               l_road_id)
      else:
        # update road ids to reflect segments with available speed data
        l_count, l_road_id = utils.generate_speed_mask(result['id'],
                                                       result['count'],
                                                       l_road_id)
    else:
      l_speed, _ = utils.generate_speed_mask(result['id'],
                                             result['speed_mph_mean'],
                                             l_samp_id)
      l_count, l_samp_id = utils.generate_speed_mask(result['id'],
                                                     result['count'], l_samp_id)

    # use updated ids to zero out invalid spots
    l_angle[l_samp_id == 0] = 0

    # normalize time (dow, hour)
    time = time.astype(float)
    time[0] = ((time[0] + 1) / 7 * 2) - 1
    time[1] = ((time[1] + 1) / 24 * 2) - 1

    t_image = TF.to_tensor(image).float()
    t_loc = torch.from_numpy(location).float()
    t_time = torch.from_numpy(time).float()
    t_road = torch.from_numpy(l_road).long()
    t_speed = torch.from_numpy(l_speed[np.newaxis, :, :]).float()
    t_angle = torch.from_numpy(l_angle[np.newaxis, :, :]).float()
    t_angle_bin = torch.from_numpy(l_angle_bin).long()
    t_count = torch.from_numpy(l_count[np.newaxis, :, :]).float()
    t_samp_id = torch.from_numpy(l_samp_id[np.newaxis, :, :]).long()
    t_road_id = torch.from_numpy(l_road_id[np.newaxis, :, :]).long()

    inputs = [t_image, t_loc, t_time]
    targets = [
        t_road, t_speed, t_angle, t_angle_bin, t_count, t_samp_id, t_road_id
    ]

    return inputs, targets

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


if __name__ == "__main__":
  take = 8

  dataset = DTSDataset('train')

  for n, data in zip(range(take), dataset):
    inputs, targets = [[y.numpy() for y in x] for x in data]
    im, loc, time = inputs
    l_road, l_speed, l_angle, l_angle_bin, _, l_samp_id, _ = targets
    print(
        n, loc.shape, time, im.shape, l_road.shape, np.quantile(im, [0, 1]),
        np.unique(l_road), np.quantile(l_angle, [0, 1]), np.unique(l_angle_bin),
        np.quantile(l_speed[l_speed > 0], [0, 1])
        if np.count_nonzero(l_speed) > 0 else [0, 0])

  dataset = DTSDataset('val')

  for n, data in zip(range(take), dataset):
    inputs, targets = [[y.numpy() for y in x] for x in data]
    im, loc, time = inputs
    l_road, l_speed, l_angle, l_angle_bin, _, l_samp_id, _ = targets
    print(
        n, loc.shape, time, im.shape, l_road.shape, np.quantile(im, [0, 1]),
        np.unique(l_road), np.quantile(l_angle, [0, 1]), np.unique(l_angle_bin),
        np.quantile(l_speed[l_speed > 0], [0, 1])
        if np.count_nonzero(l_speed) > 0 else [0, 0])