if opt_test.dataset == 'coco':
image_field = ImageDetectionsField(detections_path=os.path.join(
coco_root, 'coco_detections.hdf5'),
load_in_tmp=False)
det_field = COCOControlSequenceField(
detections_path=os.path.join(coco_root, 'coco_detections.hdf5'),
classes_path=os.path.join(coco_root, 'object_class_list.txt'),
pad_init=False,
padding_idx=-1,
all_boxes=False,
fix_length=20)
text_field = TextField(init_token='<bos>',
eos_token='<eos>',
lower=True,
remove_punctuation=True,
fix_length=20)
dataset = COCOEntities(image_field,
det_field,
text_field,
img_root='',
ann_root=os.path.join(coco_root, 'annotations'),
entities_file=os.path.join(coco_root,
'coco_entities.json'),
id_root=os.path.join(coco_root, 'annotations'))
test_dataset = COCOEntities(image_field,
det_field,
RawField(),
detections_path=os.path.join(flickr_root, 'flickr30k_detections.hdf5'),
classes_path=os.path.join(flickr_root, 'object_class_list.txt'),
img_shapes_path=os.path.join(flickr_root, 'flickr_img_shapes.json'),
precomp_glove_path=os.path.join(flickr_root, 'object_class_glove.pkl'),
verb_idx_path=os.path.join(flickr_root, 'flickr_verb_idx.json'),
idx_vs_path=os.path.join(flickr_root, 'idx_2_vs_flickr.json'),
cap_verb_path=os.path.join(flickr_root, 'cap_2_verb_flickr.json'),
cap_classes_path=os.path.join(flickr_root, 'cap_2_classes_flickr.json'),
idx_v_og_path=os.path.join(flickr_root, 'idx_2_v_og_flickr.json'),
vocab_list_path=os.path.join(flickr_root, 'vocab_tv_flickr.json'),
fix_length=10,
visual=False)
text_field = TextField(init_token='<bos>',
eos_token='<eos>',
lower=True,
remove_punctuation=True,
fix_length=20)
test_dataset = FlickrEntities(image_field,
RawField(),
det_field,
img_root='',
ann_file=os.path.join(
flickr_root, 'flickr30k_annotations.json'),
entities_root=flickr_entities_root,
verb_filter=True)
train_dataset, val_dataset, test_dataset = test_dataset.splits
test_dataset = DictionaryDataset(test_dataset.examples, test_dataset.fields,
'image')
parser.add_argument('--step_size', default=3, type=int, help='learning rate schedule step size')
parser.add_argument('--gamma', default=0.8, type=float, help='learning rate schedule gamma')
parser.add_argument('--h2_first_lstm', default=1, type=int, help='h2 as input to the first lstm')
parser.add_argument('--img_second_lstm', default=0, type=int, help='img vector as input to the second lstm')
parser.add_argument('--sample_rl', action='store_true', help='reinforcement learning with cider optimization')
parser.add_argument('--sample_rl_nw', action='store_true', help='reinforcement learning with cider + nw optimization')
opt = parser.parse_args()
print(opt)
image_field = ImageDetectionsField(detections_path=os.path.join(coco_root, 'coco_detections.hdf5'), load_in_tmp=False)
det_field = COCOControlSequenceField(detections_path=os.path.join(coco_root, 'coco_detections.hdf5'),
classes_path=os.path.join(coco_root, 'object_class_list.txt'),
pad_init=False, padding_idx=-1, all_boxes=False, fix_length=20)
text_field = TextField(init_token='<bos>', eos_token='<eos>', lower=True, remove_punctuation=True, fix_length=20)
dataset = COCOEntities(image_field, det_field, text_field,
img_root='',
ann_root=os.path.join(coco_root, 'annotations'),
entities_file=os.path.join(coco_root, 'coco_entities.json'),
id_root=os.path.join(coco_root, 'annotations'))
train_dataset, val_dataset, _ = dataset.splits
text_field.build_vocab(train_dataset, val_dataset, min_freq=5)
test_dataset = COCOEntities(image_field, det_field, RawField(),
img_root='',
ann_root=os.path.join(coco_root, 'annotations'),
entities_file=os.path.join(coco_root, 'coco_entities.json'),
id_root=os.path.join(coco_root, 'annotations'),