Smallcap: Lightweight Image Captioning Prompted with Retrieval Augmentation

Recent advances in image captioning have focused on scaling the data and model size, substantially increasing the cost of pretraining and finetuning. As an alternative to large models, we present Smallcap, which generates a caption conditioned on an input image and related captions retrieved from a datastore. Our model is lightweight and fast to train, as the only learned parameters are in newly introduced cross-attention layers between a pre-trained CLIP encoder and GPT-2 decoder. Smallcap can transfer to new domains without additional finetuning and can exploit large-scale data in a training-free fashion since the contents of the datastore can be readily replaced. Our experiments show that Smallcap, trained only on COCO, has competitive performance on this benchmark, and also transfers to other domains without retraining, solely through retrieval from target-domain data. Further improvement is achieved through the training-free exploitation of diverse human-labeled and web data, which proves to be effective for a range of domains, including the nocaps benchmark, designed to test generalization to unseen visual concepts.¹¹Code: https://github.com/RitaRamo/smallcap.

1. Introduction

The state-of-the-art in image captioning is defined by in-creasingly large-scale models trained on increasingly large-scale datasets [11], [18], [39], [42]. Scaling up leads to higher computational demands for model pretraining and finetuning on downstream tasks. This becomes especially relevant when numerous model versions may be needed for different visual domains [1] and end-users in practical applications, e.g. image captioning for the visually impaired [10].