# nocaps: novel object captioning at scale

## TL;DR

The paper introduces nocaps, a large-scale benchmark for novel object captioning that enables models to learn visual concepts from diverse data sources beyond traditional image-caption datasets.

## Contribution

It presents the first large-scale benchmark for novel object captioning at scale, combining multiple datasets and establishing strong baseline models.

## Key findings

- Benchmark includes 166,100 human-generated captions for 15,100 images.
- Nearly 400 object classes in test images have minimal or no training captions.
- Extended existing models to set baseline performance for future research.

## Abstract

Image captioning models have achieved impressive results on datasets containing limited visual concepts and large amounts of paired image-caption training data. However, if these models are to ever function in the wild, a much larger variety of visual concepts must be learned, ideally from less supervision. To encourage the development of image captioning models that can learn visual concepts from alternative data sources, such as object detection datasets, we present the first large-scale benchmark for this task. Dubbed 'nocaps', for novel object captioning at scale, our benchmark consists of 166,100 human-generated captions describing 15,100 images from the OpenImages validation and test sets. The associated training data consists of COCO image-caption pairs, plus OpenImages image-level labels and object bounding boxes. Since OpenImages contains many more classes than COCO, nearly 400 object classes seen in test images have no or very few associated training captions (hence, nocaps). We extend existing novel object captioning models to establish strong baselines for this benchmark and provide analysis to guide future work on this task.