AI Paper Atlas

Research map / Detection and segmentation / CNN features

Learning to Detect Every Thing in an Open World

Kuniaki Saito, Ping Hu, Trevor Darrell, Kate Saenko

ECCV 2022 · Detection and segmentation › CNN features · approach: Joint detection and instance segmentation · 55 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 BackErase synthetic-background augmentation with decoupled training to Mask R-CNNMask R-CNN · VOC->COCO non-VOCAR100: 10.9 -> 30.8Improved
add mask loss on real images to detection loss on synthetic images (decoupled training) versus synthetic-only training (ablation)LDET · VOC->COCO non-VOCAR100: 4.3 -> 30.8Improved
add mask loss on synthetic data to synthetic-only detection training (ablation)Synth Only · VOC->COCO non-VOCAR100: 4.3 -> 9.5Improved
add synthetic-background augmentation on top of plain real-image training (ablation)Plain Mask R-CNN · VOC->COCO non-VOCAR10: 8.8 -> 18.2Improved
reduce background crop region from 1/2 to 1/8 of image side (ablation)LDET · VOC->COCO non-VOCAR100: 26.2 -> 30.8Improved
reduce background crop region from 1/8 to 1/16 of image side (ablation)LDET · VOC->COCO non-VOCAR100: 30.8 -> 31.8Improved
reduce background crop region from 1/8 to 1/16 of image side (ablation)LDET · VOC->COCO non-VOCAR10: 18.2 -> 17.5Hurt
use DTD texture patches as external background instead of cropped self-image (ablation)LDET · VOC->COCO non-VOCbox AR: 26.7 (LDET-3.1)Hurt
add LDET to RetinaNetRetinaNet · VOC->COCO non-VOCAR100: 17.8 -> 31.0Improved
add LDET to TensorMaskTensorMask · VOC->COCO non-VOCAR100: 19.7 -> 31.1Improved
add LDET to Mask R-CNN trained on full COCOMask R-CNN · COCO->UVObox AR: 42.3 -> 47.5Improved
add LDET to Mask R-CNN trained on VOC classesMask R-CNN · COCO->UVObox AR: 30.0 -> 43.7Improved

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