dataroot=args.data_folder)
else:
val_dset = VQAFeatureDataset('val',
dictionary,
args.relation_type,
adaptive=args.adaptive,
pos_emb_dim=args.imp_pos_emb_dim,
dataroot=args.data_folder)
train_dset = VQAFeatureDataset('train',
dictionary,
args.relation_type,
adaptive=args.adaptive,
pos_emb_dim=args.imp_pos_emb_dim,
dataroot=args.data_folder)
model = build_regat(val_dset, args).to(device)
tfidf = None
weights = None
if args.tfidf:
tfidf, weights = tfidf_from_questions(['train', 'val', 'test2015'],
dictionary)
model.w_emb.init_embedding(
join(args.data_folder, 'glove/glove6b_init_300d.npy'), tfidf, weights)
model = nn.DataParallel(model).to(device)
if args.checkpoint != "":
print("Loading weights from %s" % (args.checkpoint))
if not os.path.exists(args.checkpoint):
raise ValueError("No such checkpoint exists!")
'val', model_hps.relation_type,
adaptive=model_hps.adaptive,
dataroot=model_hps.data_folder)
eval_dset = VQA_cp_Dataset(
args.split, dictionary, coco_train_features,
coco_val_features, adaptive=model_hps.adaptive,
pos_emb_dim=model_hps.imp_pos_emb_dim,
dataroot=model_hps.data_folder)
else:
eval_dset = VQAFeatureDataset(
args.split, dictionary, model_hps.relation_type,
adaptive=model_hps.adaptive,
pos_emb_dim=model_hps.imp_pos_emb_dim,
dataroot=model_hps.data_folder)
model = build_regat(eval_dset, model_hps).to(device)
model = nn.DataParallel(model).to(device)
if args.checkpoint > 0:
checkpoint_path = os.path.join(
args.output_folder,
f"model_{args.checkpoint}.pth")
else:
checkpoint_path = os.path.join(args.output_folder,
f"model.pth")
print("Loading weights from %s" % (checkpoint_path))
if not os.path.exists(checkpoint_path):
raise ValueError("No such checkpoint exists!")
checkpoint = torch.load(checkpoint_path)
state_dict = checkpoint.get('model_state', checkpoint)
action='store_true',
help='Enable bias term for relation labels \
in relation encoder')
# can use config files
parser.add_argument('--config', help='JSON config files')
args = parse_with_config(parser)
return args
if __name__ == '__main__':
args = parse_args()
n_device = torch.cuda.device_count()
print("Found %d GPU cards for training" % (n_device))
device = torch.device("cpu")
batch_size = args.batch_size * n_device
dictionary = Dictionary.load_from_file(
join(args.data_folder, 'glove/dictionary.pkl'))
val_dset = VQAFeatureDataset('val',
dictionary,
args.relation_type,
adaptive=args.adaptive,
pos_emb_dim=args.imp_pos_emb_dim,
dataroot=args.data_folder)
model = build_regat(val_dset, args)
print(model)