AI Paper Atlas

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

Learning Visual Representations with Caption Annotations

Mert Bulent Sariyildiz, Julien Perez, Diane Larlus

ECCV 2020 · Vision-language and multimodal › CLIP and zero-shot · approach: Recaptioning and text-supervised pretraining · 128 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
use 5K nouns+adjectives+verbs from captions as tags instead of 80 ground-truth labelsTP (ResNet50, COCO) · VOC07 linear SVMmAP: 80.2 -> 82.6Improved
use 5K caption tags instead of 80 ground-truth labelsTP (ResNet50, COCO) · IN-1K logistic regressiontop-1 accuracy: 34.0 -> 43.9Improved
add adjectives to noun tags (ablation)TP_postag (ResNet50) · VOC07 linear SVMmAP: 81.8 -> 82.3Improved
increase tag vocabulary from 5K to 10K (NN+ADJ+VB) (ablation)TP_postag (ResNet50) · VOC07 linear SVMmAP: 82.6 -> 81.9Hurt
replace topic-model tags (LDA) with caption-cluster tagsTP (VGG16, COCO) · VOC07 linear probe, C-13mAP: 76.3 -> 77.2Improved
use clustering of captions instead of individual noun/adjective/verb tokensTP (VGG16, COCO) · VOC07 linear probe, C-13mAP: 77.2 -> 79.4No clear effect
add masked language modeling with ICMLM_tfm on top of tag predictionTP_postag (VGG16, COCO) · VOC07 linear probe, C-13mAP: 79.4 -> 80.5Improved
add masked language modeling with ICMLM on top of tag predictionTP_postag (ResNet50, COCO) · IN-1K logistic regressiontop-1 accuracy: 43.9 -> 51.9Improved
use ICMLM_att-fc instead of ICMLM_tfmICMLM (VGG16, VG) · VOC07 linear probe, C-13mAP: 82.6 -> 85.0Improved
use ICMLM_att-fc instead of ICMLM_tfmICMLM (VGG16, VG) · IN-1K linear probe, C-13top-1 accuracy: 51.1 -> 47.8Hurt
use 2 hidden layers and 12 heads instead of 1 layer and 1 head (ablation)ICMLM_tfm (ResNet50) · VOC07 linear SVMmAP: 85.7 -> 84.8Hurt
use 2 hidden layers instead of 1 in ICMLM_tfm (ablation)ICMLM_tfm (ResNet50) · IN-1K logistic regressiontop-1 accuracy: 50.6 -> 46.6Hurt

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