AI Paper Atlas

Research map / Image generation / GANs

Towards Diverse and Natural Image Descriptions via a Conditional GAN

Bo Dai, Sanja Fidler, Raquel Urtasun, Dahua Lin

ICCV 2017 · Image generation › GANs · approach: Conditional GANs · 492 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 conditional GAN training (G-GAN) instead of MLELSTM caption generator · MSCOCO retrieval, 5000 images (E-GAN similarity)Recall@1: 5.06 -> 14.30Improved
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · MSCOCO retrieval, 5000 images (log-likelihood)Recall@1: 9.88 -> 12.04Improved
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · MSCOCO retrieval, 5000 images (log-likelihood)Recall@10: 39.94 -> 41.78Improved
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · MSCOCOCIDEr: 1.020 -> 0.795Hurt
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · Flickr30kCIDEr: 0.767 -> 0.202Hurt
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · MSCOCOE-NGAN: 0.464 -> 0.528Improved
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · MSCOCOE-GAN: 0.427 -> 0.602Improved
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · Flickr30kE-GAN: 0.439 -> 0.456Improved
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · User study pairwise preferencefraction preferred by humans: 61%Improved
add conditional GAN training (G-GAN) instead of MLELSTM caption generator · User study vs human descriptionsfraction won against human: 9% -> over 24%Improved
use E-GAN evaluator instead of CIDEr or SPICE as metriccaption evaluation · User study agreementKendall rank correlation with human evaluation: -0.30 -> 0.14Improved
add adversarial training of the evaluator (E-GAN vs E-NGAN) (ablation)caption evaluator · Human vs generator score gapdiscriminative powerImproved

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