AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Learning Object-Language Alignments for Open-Vocabulary Object Detection

Chuang Lin, Peize Sun, Yi Jiang, Ping Luo, Lizhen Qu, Gholamreza Haffari, Zehuan Yuan, Jianfei Cai

ICLR 2023 · Detection and segmentation › Open-set and few-shot detection · approach: Open-vocabulary object detection · 36 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 one-to-one Hungarian region-word matching instead of one-to-many Sinkhorn matching (ablation)VLDet region-word alignment · OV-COCOnovel mAP: +2.9 (29.1 to 32.0)Improved
Use one-to-one Hungarian region-word matching instead of one-to-many Sinkhorn matching (ablation)VLDet region-word alignment · OV-LVISmask novel AP: +3.2 (18.5 to 21.7)Improved
Enlarge object vocabulary from category names to all caption nouns (ablation)VLDet · OV-COCOnovel mAP: +1.8 (28.2 to 30.0)Improved
Enlarge object vocabulary from category names to all caption nouns (ablation)VLDet · OV-LVISmask novel AP: +1.5 (18.9 to 20.4)Improved
Add image-text alignment loss to region-word loss (ablation)VLDet · OV-COCOnovel mAP: +2.0 (30.0 to 32.0)Improved
Add image-text alignment loss to region-word loss (ablation)VLDet · OV-LVISmask novel AP: +1.3 (20.4 to 21.7)Improved
Image-text loss alone without region-word loss (ablation)VLDet · OV-COCOnovel mAP: -9.0 vs region-word only (21.0 vs 30.0)Hurt
Bipartite matching (Hungarian) with region-word loss vs. max-score/max-size assignment (ablation)VLDet region-word alignment · OV-COCO and OV-LVISnovel mAPImproved
Replace Faster R-CNN with VLDet bipartite matching on COCO Caption data vs. PB-OVDVLDet · OV-COCOnovel AP: +1.2 (30.8 to 32.0)Improved
Use Swin-B backbone instead of ResNet50VLDet · OV-LVISmask novel AP: +4.6 (21.7 to 26.3)Improved
Transfer COCO-trained VLDet without retraining to PASCAL VOC and LVISVLDet · PASCAL VOC / LVIS transferbox AP50: 61.7 vs 59.2 (PB-OVD) on VOC; 10.0 vs 8.0 on LVISImproved

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