AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

MDETR - Modulated Detection for End-to-End Multi-Modal Understanding

Aishwarya Kamath, Mannat Singh, Yann LeCun, Gabriel Synnaeve, Ishan Misra, Nicolas Carion

ICCV 2021 · Vision-language and multimodal › Human and embodied modeling · approach: Region-level vision-language alignment · 809 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 contrastive alignment loss (ablation)MDETR · CLEVR-MediumAP: 83.2 -> 99.0Improved
add soft-token classification loss (ablation)MDETR · CLEVR-MediumAP: 87.7 -> 99.0Improved
add modulated-detection pre-training curriculum (ablation)MDETR · CLEVR valQA accuracy: 68.2 -> 99.7Improved
use specialized head per question type instead of single QA head (ablation)MDETR · CLEVR valQA accuracy: 90.1 -> 99.7Improved
swap ResNet-101 backbone for EfficientNet-B3MDETR · RefCOCO valaccuracy: 86.75 -> 87.51Improved
swap EfficientNet-B3 for larger EfficientNet-B5MDETR · RefCOCO/+/gaccuracyNo clear effect
swap ResNet-101 for EfficientNet-B5MDETR · GQA test-stdaccuracy: 61.99 -> 62.45Improved
swap ResNet-101 for EfficientNet-B3MDETR · PhraseCutM-IoU: 53.1 -> 53.7Improved
add pre-training on COCO, VG and Flickr30kMDETR-R101 · Flickr30k Any-Box valR@1: 78.9 -> 82.5Improved
combine multiple annotations per image into one training textMDETR pre-training data · downstream tasksaccuracyImproved
fine-tune on CLEVR-HumansMDETR · CLEVR-Humansaccuracy: 59.9 -> 81.7Improved
train on condition A, test on condition BMDETR · CoGenTaccuracy: 99.8 -> 76.7Hurt

Later papers that compared with this method

18 later accepted papers report MDETR as a baseline in their results tables; 13 of them report a better number on most of the results they share with it.

10 more not listed.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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