BERT-hLSTMs: BERT and hierarchical LSTMs for visual storytelling

نویسندگان

چکیده

Visual storytelling is a creative and challenging task, aiming to automatically generate story-like description for sequence of images. The descriptions generated by previous visual approaches lack coherence because they use word-level generation methods do not adequately consider sentence-level dependencies. To tackle this problem, we propose novel hierarchical framework which separately models semantics. We the transformer-based BERT obtain embeddings sentences words. then employ LSTM network: bottom receives as input sentence vector representation from BERT, learn dependencies between corresponding images, top responsible generating word representations, taking LSTM. Experimental results demonstrate that our model outperforms most closely related baselines under automatic evaluation metrics BLEU CIDEr, also show effectiveness method with human evaluation.

برای دانلود باید عضویت طلایی داشته باشید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Cis Schut Bert Bredeweg

Constructing a qualitative model of some device usually proceeds as a cycle of model formulation and model debugging . The latter is driven by discrepancies between the behaviour predicted by the model and the actual device behaviour. This paper describes how the elimination of one type of discrepancy, incorrectly predicted derivatives, can be supported . It provides an analysis of the knowledg...

متن کامل

My friend Bert.

It had been a busy morning in the hospital with surgeries, rounds, and outpatient department visits. I enjoyed the few minutes of solitude as I drove to the nursing home to see my friend Bert. He sat, slouched over, in a geriatric chair and ever so slowly sat up when I spoke to him. His expressionless face could no longer welcome me, yet his trembling hands reached out to grasp mine. I sat clos...

متن کامل

Bert Bongers An Interview with

in Berlin, Edwin had the idea to form a trio. Sensorband’s first performance was in December of 1993, at Voyages Virtuels, a virtual reality exhibit organized by Les Virtualistes in Paris. I interviewed Sensorband in The Hague in November 1996, and thereafter the discussion was extended through electronic mail. The topics of the interview included: how they established an ensemble based entirel...

متن کامل

Poliovirus proves IRES - istible in vivo Bert

1678 The Journal of Clinical Investigation http://www.jci.org Volume 113 Number 12 June 2004 tional model systems and also calls for new investigations using human biological and epidemiologic data. The iterative use of human and animal studies will bring the most rapid progress toward enhanced diagnoses, interventions to improve clinical outcomes, and preventative strategies for human birth de...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

ژورنال

عنوان ژورنال: Computer Speech & Language

سال: 2021

ISSN: ['1095-8363', '0885-2308']

DOI: https://doi.org/10.1016/j.csl.2020.101169