Newer
Older
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
import os
import pdb
def mkdir_if_not_exists(dir_name):
if not os.path.exists(dir_name):
os.mkdir(dir_name)
experiment_name = 'QA_explicit_dot_joint_training_pretrained'
#experiment_name = 'object_attribute_classifier_large_images'
# Global output directory (all subexperiments will be saved here)
global_output_dir = '/home/tanmay/Code/GenVQA/Exp_Results/VQA'
global_experiment_dir = os.path.join(
global_output_dir,
experiment_name)
tb_log_dir = os.path.join(
global_experiment_dir,
'tensorboard_logdir')
mkdir_if_not_exists(global_output_dir)
mkdir_if_not_exists(global_experiment_dir)
mkdir_if_not_exists(tb_log_dir)
#height and width to which images are resized before feeding into networks
image_size = (224, 224)
# Token to be used if object or attribute variable is unknown
unknown_token = 'UNK'
# Genome Data paths
data_absolute_path = '/home/ssd/VisualGenome'
image_dir = os.path.join(data_absolute_path, 'cropped_regions_large')
genome_resnet_feat_dir = os.path.join(
data_absolute_path,
'cropped_regions_large_resnet_features')
object_labels_json = os.path.join(
data_absolute_path,
'restructured/object_labels.json')
attribute_labels_json = os.path.join(
data_absolute_path,
'restructured/attribute_labels.json')
regions_json = os.path.join(
data_absolute_path,
'restructured/region_with_labels.json')
mean_image_filename = os.path.join(
data_absolute_path,
'restructured/mean_image.jpg')
vocab_json = os.path.join(
data_absolute_path,
'restructured/vocab_subset.json')
num_object_labels = 1000
num_attribute_labels = 1000
# Regions data partition
# First 80% meant to be used for training
# Next 10% is set aside for validation
# Last 10% is to be used for testing
num_total_regions = 1951768
num_train_regions = 1561416 # First 80%
num_val_regions = 195176 # Next 10%
num_test_regions = num_total_regions \
- num_train_regions \
- num_val_regions
# Pretrained resnet ckpt
resnet_ckpt = '/home/tanmay/Downloads/pretrained_networks/' + \
'Resnet/tensorflow-resnet-pretrained-20160509/' + \
'ResNet-L50.ckpt'
# Pretrained word vectors
word2vec_binary = '/home/tanmay/Code/word2vec/word2vec-api-master/' + \
'GoogleNews-vectors-negative300.bin'
word_vector_size = 300
resnet_feat_dim = 2048
# Numpy matrix storing vocabulary word vectors
pretrained_vocab_word_vectors_npy = os.path.join(
data_absolute_path,
'restructured/pretrained_vocab_word_vectors.npy')
# Object Attribute Classifier Training Params
region_batch_size = 200
region_num_samples = num_train_regions
region_num_epochs = 10
region_offset = 0
region_queue_size = 400
region_regularization_coeff = 1e-4
region_lr = 1e-4
region_log_every_n_iter = 500
region_output_dir = os.path.join(
global_experiment_dir,
'object_attribute_classifiers')
mkdir_if_not_exists(region_output_dir)
region_model = os.path.join(
region_output_dir,
'model')
# Object Attribute Finetuning Params
region_fine_tune_from_iter = 3000
region_fine_tune_from = region_model + '-' + str(region_fine_tune_from_iter)
# Object Attribute Classifier Evaluation Params
region_eval_on = 'train' # One of {'val','test','train'}
region_model_to_eval = region_model + '-' + '77500'
region_attribute_scores_dirname = os.path.join(
region_output_dir,
'attribute_scores')
mkdir_if_not_exists(region_attribute_scores_dirname)
# Answer prediction
num_region_proposals = 100
num_mcq_candidates = 18
num_negative_answers = num_mcq_candidates - 1
# VQA data paths
vqa_basedir = '/home/ssd/VQA/'
vqa_train_image_dir = os.path.join(
vqa_basedir,
'train2014_cropped_large')
vqa_train_resnet_feat_dir = os.path.join(
vqa_basedir,
'train2014_cropped_large_resnet_features')
vqa_train_anno = os.path.join(
vqa_basedir,
'mscoco_train2014_annotations_with_parsed_questions.json')
vqa_val_image_dir = os.path.join(
vqa_basedir,
'val2014_cropped_large')
vqa_val_anno = os.path.join(
vqa_basedir,
'mscoco_val2014_annotations_with_parsed_questions.json')
vqa_answer_vocab_json = os.path.join(
vqa_basedir,
'answer_vocab.json')
# VQA dataset params
num_train_questions = 248349
num_val_questions = 10000 #121512
num_test_questions = 0
# Answer classifier training params
answer_batch_size = 50
answer_num_epochs = 10
answer_offset = 0
answer_regularization_coeff = 1e-5
answer_queue_size = 500
answer_embedding_dim = 600
answer_lr = 1e-4
answer_log_every_n_iter = 500
answer_output_dir = os.path.join(
global_experiment_dir,
'answer_classifiers')
mkdir_if_not_exists(answer_output_dir)
pretrained_model = '/home/tanmay/Code/GenVQA/Exp_Results/VisualGenome/' + \
'object_attribute_classifier_large_images/' + \
'object_attribute_classifiers/model-77500'
answer_model = os.path.join(
answer_output_dir,
'model')
# Answer classifier additional joint training params
num_regions_with_labels = 100
# Answer fine tune params
answer_fine_tune_from_iter = 17000
answer_fine_tune_from = answer_model + '-' + str(answer_fine_tune_from_iter)
# Answer eval params
answer_eval_on = 'val'
answer_model_to_eval = answer_model + '-13000'
answer_eval_data_json = os.path.join(
answer_output_dir,
'eval_' + answer_eval_on + '_data.json')