Reproduce “Length-Controllable Image Captioning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chaorui Deng, Ning Ding, Mingkui Tan, Qi Wu
Researchers introduce methods for length-controllable image captioning, allowing users to specify desired caption lengths. This work proposes a length level embedding that integrates into existing autoregressive models and presents LaBERT, a novel non-autoregressive model, achieving competitive quality with enhanced decoding efficiency and diversity, particularly for longer captions.
Key insights
- A simple, plug-and-play length level embedding can effectively enable explicit length control in diverse autoregressive captioning architectures (LSTM and Transformer) without compromising caption quality.
- Non-autoregressive generation through iterative refinement can achieve competitive caption quality while offering substantial decoding speedups (up to 2.1x for long captions) compared to autoregressive models.
- Non-autoregressive models, like LaBERT, intrinsically foster greater diversity in generated captions, significantly outperforming autoregressive methods on standard diversity metrics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.