AI Paper Atlas

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

Generative Negative Text Replay for Continual Vision-Language Pretraining

Shipeng Yan, Lanqing Hong, Hang Xu, Jianhua Han, Tinne Tuytelaars, Zhenguo Li, Xuming He

ECCV 2022 · Vision-language and multimodal › CLIP and zero-shot · approach: Self-distillation for vision-language pretraining · 22 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 knowledge distillation loss (ablation)IncCLIP · CC2M class incremental (4 steps), zero-shot ImageNettop-1 accuracy: 16.01 -> 17.67Improved
add hard negative text generation (ablation)IncCLIP · CC2M class incremental (4 steps), zero-shot ImageNettop-1 accuracy: 17.67 -> 18.85Improved
add hard negative text generation (ablation)IncCLIP · Flickr30K image-to-text (class incremental, 4 steps)R@1: 22.8 -> 24.1Improved
add knowledge distillation loss (ablation)IncCLIP · Flickr30K image-to-text (class incremental, 4 steps)R@1: 21.3 -> 22.8Improved
add hard negative text generation (ablation)IncCLIP · Flickr30K text-to-image (class incremental, 4 steps)R@1: 15.10 -> 17.14Improved
add hard negative text generation (ablation)IncCLIP · MSCOCO image-to-text (class incremental, 4 steps)R@1: 11.55 -> 12.38Improved
add weight norm clipping (ablation)IncCLIP · CC2M class incremental (4 steps), zero-shot ImageNettop-1 accuracyImproved
increase memory size from 1% to 50% of dataset (ablation)IncCLIP · CC2M class incremental (4 steps), zero-shot ImageNettop-1 accuracy: 12.82 -> 28.29Improved
increase memory size from 1% to 50% of dataset (ablation)IncCLIP · Flickr30K image-to-text (class incremental, 4 steps)R@1: 18.2 -> 31.6Improved
use IncCLIP instead of Co2Lclass incremental 4 steps · CC2M class incremental (4 steps), average of 8 datasetstop-1 accuracy: 16.84 -> 21.57Improved
use IncCLIP instead of ER for forgettingcontinual VLP · CC2M class incremental, forgetting analysisBWT: -31.93 -> -18.48Improved
increase number of steps from 4 to 8IncCLIP · CC2M class incremental, average of 8 datasetstop-1 accuracy: 21.57 -> 17.24Hurt

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