File size: 8,127 Bytes
a166479
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
import os
import sys
import torch.utils.data as data
import torch
from torchvision import transforms
from torch.autograd import Variable
import numpy as np
from PIL import Image
import torchvision.transforms.functional as TF
import random

from bert.tokenization_bert import BertTokenizer

import h5py
from refer.refer import REFER

from args import get_parser

# Dataset configuration initialization
parser = get_parser()
args = parser.parse_args()

from hfai.datasets import CocoDetection

from PIL import Image
import numpy as np
#from ffrecord.torch import DataLoader,Dataset
import ffrecord

_EXIF_ORIENT = 274


from itertools import permutations

def _apply_exif_orientation(image):
    """
    Applies the exif orientation correctly.

    This code exists per the bug:
      https://github.com/python-pillow/Pillow/issues/3973
    with the function `ImageOps.exif_transpose`. The Pillow source raises errors with
    various methods, especially `tobytes`

    Function based on:
      https://github.com/wkentaro/labelme/blob/v4.5.4/labelme/utils/image.py#L59
      https://github.com/python-pillow/Pillow/blob/7.1.2/src/PIL/ImageOps.py#L527

    Args:
        image (PIL.Image): a PIL image

    Returns:
        (PIL.Image): the PIL image with exif orientation applied, if applicable
    """
    if not hasattr(image, "getexif"):
        return image

    try:
        exif = image.getexif()
    except Exception:  # https://github.com/facebookresearch/detectron2/issues/1885
        exif = None

    if exif is None:
        return image

    orientation = exif.get(_EXIF_ORIENT)

    method = {
        2: Image.FLIP_LEFT_RIGHT,
        3: Image.ROTATE_180,
        4: Image.FLIP_TOP_BOTTOM,
        5: Image.TRANSPOSE,
        6: Image.ROTATE_270,
        7: Image.TRANSVERSE,
        8: Image.ROTATE_90,
    }.get(orientation)

    if method is not None:
        return image.transpose(method)
    return image

def convert_PIL_to_numpy(image, format):
    """
    Convert PIL image to numpy array of target format.

    Args:
        image (PIL.Image): a PIL image
        format (str): the format of output image

    Returns:
        (np.ndarray): also see `read_image`
    """
    if format is not None:
        # PIL only supports RGB, so convert to RGB and flip channels over below
        conversion_format = format
        if format in ["BGR", "YUV-BT.601"]:
            conversion_format = "RGB"
        image = image.convert(conversion_format)
    image = np.asarray(image)
    # PIL squeezes out the channel dimension for "L", so make it HWC
    if format == "L":
        image = np.expand_dims(image, -1)

    # handle formats not supported by PIL
    elif format == "BGR":
        # flip channels if needed
        image = image[:, :, ::-1]
    elif format == "YUV-BT.601":
        image = image / 255.0
        image = np.dot(image, np.array(_M_RGB2YUV).T)

    return image

class ReferDataset(data.Dataset):
#class ReferDataset(ffrecord.torch.Dataset):

    def __init__(self,
                 args,
                 image_transforms=None,
                 target_transforms=None,
                 split='train',
                 eval_mode=False):

        self.classes = []
        self.image_transforms = image_transforms
        self.target_transform = target_transforms
        self.split = split
        self.refer = REFER(args.refer_data_root, args.dataset, args.splitBy)

        #self.max_tokens = 20
        self.max_tokens = 40

        ref_ids = self.refer.getRefIds(split=self.split)
        img_ids = self.refer.getImgIds(ref_ids)

        all_imgs = self.refer.Imgs
        self.imgs = list(all_imgs[i] for i in img_ids)
        self.ref_ids = ref_ids

        self.input_ids = []
        self.attention_masks = []
        self.tokenizer = BertTokenizer.from_pretrained(args.bert_tokenizer)

        self.eval_mode = eval_mode
        # if we are testing on a dataset, test all sentences of an object;
        # o/w, we are validating during training, randomly sample one sentence for efficiency
        for r in ref_ids:
            ref = self.refer.Refs[r]

            sentences_for_ref = []
            attentions_for_ref = []

            # initializing list
            test_list = ref['sentences']
             
            # All possible concatenations in String List
            # Using permutations() + loop
            if not self.eval_mode:
                test_list = [el['raw'] for el in test_list]
                res = []

                for k in test_list:
                    for l in test_list:
                        res.append(k + ', ' + l)
                #for idx in range(1, len(test_list) + 1):
                #    temp.extend(list(permutations(test_list, idx)))
                #res = []
                #for ele in temp:
                #    res.append(", ".join(ele))
            else:
                print("eval mode")
                test_list = [el['raw'] + ', ' + el ['raw'] for el in test_list]
                res = test_list

            #for i, (el, sent_id) in enumerate(zip(ref['sentences'], ref['sent_ids'])):
            for i, el in enumerate(res):
                sentence_raw = el
                attention_mask = [0] * self.max_tokens
                padded_input_ids = [0] * self.max_tokens

                input_ids = self.tokenizer.encode(text=sentence_raw, add_special_tokens=True)

                # truncation of tokens
                input_ids = input_ids[:self.max_tokens]

                padded_input_ids[:len(input_ids)] = input_ids
                attention_mask[:len(input_ids)] = [1]*len(input_ids)

                sentences_for_ref.append(torch.tensor(padded_input_ids).unsqueeze(0))
                attentions_for_ref.append(torch.tensor(attention_mask).unsqueeze(0))

            self.input_ids.append(sentences_for_ref)
            self.attention_masks.append(attentions_for_ref)

        split = 'train'
        print(split)
        self.hfai_dataset = CocoDetection(split, transform=None)
        self.keys = {}
        for i in range(len(self.hfai_dataset.reader.ids)):
            self.keys[self.hfai_dataset.reader.ids[i]] = i

    def get_classes(self):
        return self.classes

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

    def __getitem__(self, index):
        #print(index)
        #index = index[0]
        this_ref_id = self.ref_ids[index]
        this_img_id = self.refer.getImgIds(this_ref_id)
        this_img = self.refer.Imgs[this_img_id[0]]

        #print("this_ref_id", this_ref_id)
        #print("this_img_id", this_img_id)
        #print("this_img", this_img)
        #img = Image.open(os.path.join(self.refer.IMAGE_DIR, this_img['file_name'])).convert("RGB")
        img = self.hfai_dataset.reader.read_imgs([self.keys[this_img_id[0]]])[0]
        img = _apply_exif_orientation(img)
        img = convert_PIL_to_numpy(img, 'RGB')
        #print(img.shape)
        img = Image.fromarray(img)

        ref = self.refer.loadRefs(this_ref_id)

        ref_mask = np.array(self.refer.getMask(ref[0])['mask'])
        annot = np.zeros(ref_mask.shape)
        annot[ref_mask == 1] = 1

        annot = Image.fromarray(annot.astype(np.uint8), mode="P")

        if self.image_transforms is not None:
            # resize, from PIL to tensor, and mean and std normalization
            img, target = self.image_transforms(img, annot)

        if self.eval_mode:
            embedding = []
            att = []
            for s in range(len(self.input_ids[index])):
                e = self.input_ids[index][s]
                a = self.attention_masks[index][s]
                embedding.append(e.unsqueeze(-1))
                att.append(a.unsqueeze(-1))

            tensor_embeddings = torch.cat(embedding, dim=-1)
            attention_mask = torch.cat(att, dim=-1)
        else:
            choice_sent = np.random.choice(len(self.input_ids[index]))
            tensor_embeddings = self.input_ids[index][choice_sent]
            attention_mask = self.attention_masks[index][choice_sent]

        #print(img.shape)
        return img, target, tensor_embeddings, attention_mask