Face_Recognition/DatasetLoader.py at main · nitnizzie/Face_Recognition · 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
#! /usr/bin/python
# -*- encoding: utf-8 -*-

import torch
import numpy
import random
import math
import pdb
import glob
import os
from torch.utils.data import Dataset, DataLoader
from PIL import Image

def round_down(num, divisor):
    return num - (num%divisor)

def worker_init_fn(worker_id):
    numpy.random.seed(numpy.random.get_state()[1][0] + worker_id)

class meta_loader(Dataset):
    def __init__(self, train_path, train_ext, transform):

        ## Read Training Files
        files = glob.glob('%s/*/*.%s'%(train_path,train_ext))

        ## Make a mapping from Class Name to Class Number
        dictkeys = list(set([x.split('/')[-2] for x in files]))
        dictkeys.sort()
        dictkeys = { key : ii for ii, key in enumerate(dictkeys) }

        self.transform  = transform

        self.label_dict = {}
        self.data_list  = []
        self.data_label = []

        for lidx, file in enumerate(files):
            speaker_name = file.split('/')[-2]
            speaker_label = dictkeys[speaker_name];

            if not (speaker_label in self.label_dict):
                self.label_dict[speaker_label] = [];

            self.label_dict[speaker_label].append(lidx);

            self.data_label.append(speaker_label)
            self.data_list.append(file)

        print('{:d} files from {:d} classes found.'.format(len(self.data_list),len(self.label_dict)))

    def __getitem__(self, indices):

        feat = []
        for index in indices:
            feat.append(self.transform(Image.open(self.data_list[index])));
        feat = numpy.stack(feat, axis=0)

        return torch.FloatTensor(feat), self.data_label[index]

    def __len__(self):

        return len(self.data_list)

class test_dataset_loader(Dataset):
    def __init__(self, test_list, test_path, transform, **kwargs):
        self.test_path  = test_path
        self.data_list  = test_list
        self.transform  = transform

    def __getitem__(self, index):
        img = Image.open(os.path.join(self.test_path, self.data_list[index]))
        return self.transform(img), self.data_list[index]

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


class meta_sampler(torch.utils.data.Sampler):
    def __init__(self, data_source, nPerClass, max_img_per_cls, batch_size):

        self.label_dict         = data_source.label_dict
        self.nPerClass          = nPerClass
        self.max_img_per_cls    = max_img_per_cls;
        self.batch_size         = batch_size;

        self.num_iters          = 0

    def __iter__(self):

        ## Get a list of identities
        dictkeys = list(self.label_dict.keys());
        dictkeys.sort()

        lol = lambda lst, sz: [lst[i:i+sz] for i in range(0, len(lst), sz)]

        flattened_list = []
        flattened_label = []

        ## Data for each class
        for findex, key in enumerate(dictkeys):
            data    = self.label_dict[key]
            numSeg  = round_down(min(len(data),self.max_img_per_cls),self.nPerClass)

            rp      = lol(numpy.random.permutation(len(data))[:numSeg],self.nPerClass)
            flattened_label.extend([findex] * (len(rp)))
            for indices in rp:
                flattened_list.append([data[i] for i in indices])

        ## Data in random order
        mixid           = numpy.random.permutation(len(flattened_label))
        mixlabel        = []
        mixmap          = []

        ## Prevent two pairs of the same speaker in the same batch
        for ii in mixid:
            startbatch = len(mixlabel) - len(mixlabel) % self.batch_size
            if flattened_label[ii] not in mixlabel[startbatch:]:
                mixlabel.append(flattened_label[ii])
                mixmap.append(ii)

        batch_indices = [flattened_list[i] for i in mixmap]

        self.num_iters = len(batch_indices)

        return iter(batch_indices)

    def __len__(self):
        return self.num_iters

def get_data_loader(batch_size, max_img_per_cls, nDataLoaderThread, nPerClass, train_path, train_ext, transform, **kwargs):

    train_dataset = meta_loader(train_path, train_ext, transform)

    train_sampler = meta_sampler(train_dataset, nPerClass, max_img_per_cls, batch_size)

    train_loader = torch.utils.data.DataLoader(
        train_dataset,
        batch_size=batch_size,
        num_workers=nDataLoaderThread,
        sampler=train_sampler,
        pin_memory=False,
        worker_init_fn=worker_init_fn,
        drop_last=True,
    )

    return train_loader