AI Paper Atlas

Research map / Detection and segmentation / CNN features

Lite-MDETR: A Lightweight Multi-Modal Detector

Qian Lou, Yen-Chang Hsu, Burak Uzkent, Ting Hua, Yilin Shen, Hongxia Jin

CVPR 2022 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 7 citations (OpenAlex)

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
Replace text-encoder linear transformations with DLT layersMDETR-ENB3 · RefCOCOaccuracy: 87.5 -> 85.4Hurt
Replace text-encoder and multi-modal-transformer linear transformations with DLT layersMDETR-ENB3 · RefCOCOaccuracy: 87.5 -> 85.1Hurt
Add 16-bit orthogonal quantization to Lite-MDETR-TT (ablation)Lite-MDETR-TT · RefCOCO+accuracy: 80.5 -> 80.1Hurt
Replace text-encoder and multi-modal-transformer linear transformations with DLT layersMDETR-ENB3 · PhraseCutmean-IoU: 53.7 -> 49.7Hurt
Apply DLT to the text and multi-modal transformer and 16-bit quantizationMDETR-ENB5 · GQA test-devaccuracy: 62.95 -> 59.6Hurt
Increase the text-encoder dictionary ratio from 20% to 60% (ablation)Lite-MDETR-TTQ · RefCOCOaccuracy: 83.2 ±0.3 -> 85.4±0.3Improved
Increase the text-encoder dictionary ratio from 20% to 60% (ablation)Lite-MDETR-TTQ · PhraseCutmean-IoU: 46.5±0.3 -> 50.3±0.2Improved
Increase the text-encoder dictionary ratio from 20% to 60% (ablation)Lite-MDETR-TTQ · Flickr30K MergedBoxR@1: 75.6±0.2 -> 80.7±0.3Improved
Use a larger text-encoder dictionary ratio (ablation)Lite-MDETR-TTQ · RefCOCOmodel size: 80MB -> 171MBHurt

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