-
Notifications
You must be signed in to change notification settings - Fork 2
/
makeDatasetRGB.py
169 lines (148 loc) · 6.74 KB
/
makeDatasetRGB.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
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
168
import os
import torch
from torch.utils.data import Dataset
from PIL import Image
import numpy as np
import glob
import random
import re
def gen_split(root_dir, stackSize):
Dataset = []
Labels = []
NumFrames = []
root_dir = os.path.join(root_dir, 'frames')
for dir_user in sorted(os.listdir(root_dir)):
class_id = 0
dir = os.path.join(root_dir, dir_user)
for target in sorted(os.listdir(dir)):
dir1 = os.path.join(dir, target)
insts = sorted(os.listdir(dir1))
if insts != []:
for inst in insts:
inst_dir = os.path.join(dir1, inst)
numFrames = len(glob.glob1(inst_dir, '*.jpg'))
if numFrames >= stackSize:
Dataset.append(inst_dir)
Labels.append(class_id)
NumFrames.append(numFrames)
class_id += 1
return Dataset, Labels, NumFrames
class makeDataset(Dataset):
def __init__(self, root_dir, spatial_transform=None, seqLen=20,
train=True, mulSeg=False, numSeg=1, fmt='.jpg'):
self.images, self.labels, self.numFrames = gen_split(root_dir, 5)
self.spatial_transform = spatial_transform
self.train = train
self.mulSeg = mulSeg
self.numSeg = numSeg
self.seqLen = seqLen
self.fmt = fmt
def __len__(self):
return len(self.images)
def __getitem__(self, idx):
vid_name = self.images[idx]
label = self.labels[idx]
numFrame = self.numFrames[idx]
inpSeq = []
self.spatial_transform.randomize_parameters()
for i in np.linspace(1, numFrame, self.seqLen, endpoint=False):
fl_name = vid_name + '/' + 'image_' + str(int(np.floor(i))).zfill(5) + self.fmt
img = Image.open(fl_name)
inpSeq.append(self.spatial_transform(img.convert('RGB')))
inpSeq = torch.stack(inpSeq, 0)
return inpSeq, label
def gen_split_supervision(root_dir):
Dataset = []
Labels = []
NumFrames = []
data_mmaps=[]
root_dir = os.path.join(root_dir, 'frames')
for dir_user in sorted(os.listdir(root_dir)):
class_id = 0
dir = os.path.join(root_dir, dir_user)
for target in sorted(os.listdir(dir)):
dir1 = os.path.join(dir, target)
insts = sorted(os.listdir(dir1))
if insts != []:
for inst in insts:
inst_dir = os.path.join(dir1, inst, "rgb")
if os.path.isdir(inst_dir)==False:
continue
inst_dir_mmaps = os.path.join(dir1, inst, "mmaps")
numFrames_mmaps = len(glob.glob1(inst_dir_mmaps, '*.png'))
numFrames = len(glob.glob1(inst_dir, '*.png'))
Dataset.append(inst_dir)
Labels.append(class_id)
NumFrames.append(numFrames)
data_mmaps.append(inst_dir_mmaps)
class_id += 1
return Dataset, Labels, NumFrames, data_mmaps
class MakeDataset_flowsupervision(Dataset):
def __init__(self, root_dir, spatial_transform=None ,seq_len=20,
train=True, mulSeg=False, numSeg=1, fmt='.png'):
self.images, self.labels, self.numFrames,self.mmaps = gen_split_supervision(root_dir)
self.spatial_transform = spatial_transform
self.train = train
self.mulSeg = mulSeg
self.numSeg = numSeg
self.seq_len = seq_len
self.fmt = fmt
def __len__(self):
return len(self.images)
def __getitem__(self, idx):
vid_name = self.images[idx]
label = self.labels[idx]
mmaps_name = self.mmaps[idx]
numFrame = self.numFrames[idx]
inpSeq = []
m=[]
self.spatial_transform.randomize_parameters()
for i in np.linspace(1, numFrame, self.seq_len, endpoint=False):
fl_name = vid_name + '/' + 'rgb' + str(int(np.floor(i))).zfill(4) + self.fmt
fl_name_mmaps = vid_name + '/' + 'rgb' + str(int(np.floor(i+1))).zfill(4) + self.fmt
if (not os.path.isfile(fl_name_mmaps)):#in case is not present the a element of the self-supervised task we put the flag=0
fl_name_mmaps = vid_name + '/' + 'rgb' + str(int(np.floor(i))).zfill(4) + self.fmt
img = Image.open(fl_name)
img_maps = Image.open(fl_name_mmaps)
inpSeq.append(self.spatial_transform(img.convert('RGB')))
m.append(self.spatial_transform(img_maps.convert('RGB')))
inpSeq = torch.stack(inpSeq, 0)
m = torch.stack(m, 0)
return inpSeq, label, m
class makeDataset_supervision(Dataset):
def __init__(self, root_dir, spatial_transform=None, spatial_transform_map=None,seqLen=20,
train=True, mulSeg=False, numSeg=1, fmt='.png'):
self.images, self.labels, self.numFrames,self.mmaps = gen_split_supervision(root_dir)
self.spatial_transform = spatial_transform
self.spatial_transform_map = spatial_transform_map
self.train = train
self.mulSeg = mulSeg
self.numSeg = numSeg
self.seqLen = seqLen
self.fmt = fmt
def __len__(self):
return len(self.images)
def __getitem__(self, idx):
vid_name = self.images[idx]
mmaps_name= self.mmaps[idx]
label = self.labels[idx]
numFrame = self.numFrames[idx]
inpSeq = []
mmapsSeq=[]
self.spatial_transform.randomize_parameters()
if self.train==True:#applies the same transforms also at the target of the self-supervised task (eliminating random elements)
self.spatial_transform_map.transforms[2].scale=self.spatial_transform.transforms[2].scale
self.spatial_transform_map.transforms[2].crop_position=self.spatial_transform.transforms[2].crop_position
self.spatial_transform_map.transforms[1].p=self.spatial_transform.transforms[1].p
for i in np.linspace(1, numFrame, self.seqLen, endpoint=False):
fl_name = vid_name + '/' + 'rgb' + str(int(np.floor(i))).zfill(4) + self.fmt
fl_name_mmaps = mmaps_name + '/' + 'map' + str(int(np.floor(i))).zfill(4) + self.fmt
if (not os.path.isfile(fl_name_mmaps)):#in case is not present the a element of the self-supervised task we put the flag=0
fl_name_mmaps = mmaps_name + '/' + 'map' + str(int(np.floor(i+1))).zfill(4) + self.fmt
img = Image.open(fl_name)
img_maps = Image.open(fl_name_mmaps)
inpSeq.append(self.spatial_transform(img.convert('RGB')))
mmapsSeq.append(self.spatial_transform_map(img_maps ))
inpSeq = torch.stack(inpSeq, 0)
mmapsSeq=torch.stack(mmapsSeq, 0)
return inpSeq, label,mmapsSeq