Self-Distillation for Few-Shot Image Captioning (Supplementary Materials)

In this supplementary material, we provide additional information and results of our proposed few-shot image captioner on the Microsoft COCO dataset [7] and Flickr30K dataset [8]. In Section 1, we present the algorithm of our few-shot image captioning. Next, in Section 2, we show a comprehensive ablation study for the hyper-parameters and the effects on different base models. Further, in Section 3, we present the quantitative results for Flickr30K [8] with state-of-the-art approach and our different baselines. Finally, in Section 4, we report and discuss qualitative results of various baselines and state-of-the-art approaches.