AI Paper Atlas

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

Detecting Twenty-Thousand Classes Using Image-Level Supervision

Xingyi Zhou, Rohit Girdhar, Armand Joulin, Philipp Krähenbühl, Ishan Misra

ECCV 2022 · Detection and segmentation › Open-set and few-shot detection · approach: Open-set and unknown object detection · 660 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 Detic max-size image-label loss (IN-L)Box-Supervised · Open-vocab LVISmask mAP novel: 16.3 -> 24.6Improved
add Detic max-size image-label loss (CC)Box-Supervised · Open-vocab LVISmask mAP novel: 16.3 -> 19.5Improved
use max-size non-prediction loss instead of YOLO9000 prediction-based assignment (IN-L) (ablation)Detic · Open-vocab LVISmask mAP novel: 20.4 -> 24.6Improved
use prediction-based self-training with image labels (ablation)Box-Supervised · Open-vocab LVISmask mAP novel: 16.3 -> 15.6Hurt
use Detic with ViLD training setupViLD-text · Open-vocab LVIS (ViLD setting)mask mAP novel: 10.1 -> 17.8Improved
add Detic to OVR-CNN setupBase-only (CLIP) · Open-vocab COCObox mAP50 novel: 1.3 -> 27.8Improved
train with all 21K ImageNet classes instead of IN-L (ablation)Detic · Objects365box mAP rare: 17.8 -> 20.0Improved
train with all 21K classes instead of IN-L (ablation)Detic · OpenImagesbox mAP50: 53.0 -> 55.2Improved
add Detic with trained (random init) classifier (ablation)Box-Supervised · Open-vocab LVISmask mAP novel: 0 -> 17.4Improved
replace CLIP classifier with FastText (ablation)Detic · Open-vocab LVISmask mAP novel: 24.9 -> 19.2Hurt
add Detic with IN-1K pretraining (ablation)Box-Supervised · Open-vocab LVISmask mAP novel: 13.6 -> 21.7Improved
add Detic image-level labels on standard LVISBaseline Swin-B · Standard LVISmask mAP rare: 35.9 -> 41.7Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

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