Question about Results

[WangLanxiao]

Thanks for your work and code. It is a pretty good paper.

Could you share the best model of text-only image captioning on MSCOCO or the generated results json for 5000 images of MSCOCO test.
I want to add some visualization of your method.

Thanks for your help.

[yangbang18]

Thanks for your interest. Here I provide the in-domain zero-shot caption results.

Attachment: in_domain_zero_shot_coco.json

Detailed scores of this file are:
Bleu_1: 0.692
Bleu_2: 0.515
Bleu_3: 0.378
Bleu_4: 0.276
METEOR: 0.252
ROUGE_L: 0.516
CIDEr: 0.962
SPICE: 0.185