AI Paper Atlas

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

Open-Vocabulary Attribute Detection

María A. Bravo, Sudhanshu Mittal, Simon Ging, Thomas Brox

CVPR 2023 · Detection and segmentation › Open-set and few-shot detection · approach: Open-vocabulary object detection · 29 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
Train using caption and parts-of-caption proxy-labels, including noun complements, rather than only base-class boxes (ablation)OVAD-Baseline · OVADmAP: 11.7\std{0.1} -> 18.8\std{0.3}Improved
Add caption supervision (ablation)OVAD-Baseline · OVADmAP: 11.7\std{0.1} -> 15.0\std{0.2}Improved
Add noun proxy-labels to caption supervision (ablation)OVAD-Baseline · OVADmAP: 15.0\std{0.2} -> 16.2\std{0.3}Improved
Use noun phrases instead of noun complements as additional proxy-labels (ablation)OVAD-Baseline · OVADmAP: 18.8\std{0.3} -> 15.9\std{0.1}Hurt
Add noun-complement proxy-labels (ablation)OVAD-Baseline · OVADmAP: 16.2\std{0.3} -> 18.8\std{0.3}Improved
Scale OpenCLIP training data from 400M to 2B image-text pairsOpen CLIP · OVAD-BoxmAP (All): 16.0 -> 17.0; 6.25%Improved
Scale BLIP training data from 14M to 129MBLIP · OVAD-BoxmAP (All): 17.0 -> 18.2; 7.06%Improved
Fine-tune ALBEF on curated RefCOCO+ captionsALBEF · OVAD-BoxmAP (All): 15.3 -> 21.0; 37.25%Improved
Fine-tune BLIP on COCO captionsBLIP · OVAD-BoxmAP (All): 18.2 -> 24.3; 33.52%Improved
Fine-tune X-VLM for image-caption retrievalX-VLM · OVAD-BoxmAP (All): 28.1 -> 26.2; 6.76%Hurt
Train with localized region-text matchingX-VLM · OVAD-BoxmAP (All): 28.1Improved
Train a supervised attribute detector on OVAD training foldsResNet50 attribute detector · OVAD-BoxmAP: 8.29{ $\pm$0.06} -> 48.16{ $\pm$0.52}Improved

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