AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Towards Unsupervised Image Captioning With Shared Multimodal Embeddings

Iro Laina, Christian Rupprecht, Nassir Navab

ICCV 2019 · Vision-language and multimodal › CLIP and zero-shot · approach: Recaptioning and text-supervised pretraining · 108 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add robust min-distance alignment loss instead of L2 (ablation)Joint captioning model · COCO test (unpaired, 80 COCO categories)CIDEr: 53.8 -> 62.4Improved
add naive L2 alignment to MLE training (ablation)MLE-only model · COCO test (unpaired, 80 COCO categories)CIDEr: 57.9 -> 53.8No clear effect
add triplet loss to the language model (ablation)Joint robust model · COCO test (unpaired, 80 COCO categories)CIDEr: 55.6 -> 62.4Improved
add adversarial training conditioned on visual concepts (ablation)Joint robust model · COCO test (unpaired, 80 COCO categories)CIDEr: 62.4 -> 63.6Improved
add adversarial training conditioned on visual concepts (ablation)Joint robust model · COCO test (unpaired, 80 COCO categories)BLEU-4: 18.8 -> 19.3Improved
add adversarial training (ablation)Unsupervised model · Flickr30k images / COCO captions (unsupervised)CIDEr: 8.2 -> 9.9Improved
add adversarial training (ablation)Unsupervised model · COCO images / Conceptual Captions (unsupervised)CIDEr: 20.8 -> 22.7Improved
add adversarial training (ablation)Unsupervised model · Flickr30k images / COCO captions (unsupervised)BLEU-4: 5.9 -> 7.9Improved
add adversarial training (ablation)Unsupervised model · COCO images / Conceptual Captions (unsupervised)BLEU-4: 5.5 -> 6.5Improved
jointly finetune decoder instead of freezing it (ablation)Alignment-only model · COCO test (unpaired, 80 COCO categories)CIDEr: 29.4 -> 57.9Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy