AI Paper Atlas

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

"A Simple Approach and Benchmark for 21,000-Category Object Detection"

Yutong Lin, Chen Li, Yue Cao, Zheng Zhang, Jianfeng Wang, Lijuan Wang, Zicheng Liu, Han Hu

ECCV 2022 · Detection and segmentation › Open-set and few-shot detection · approach: Open-set and unknown object detection

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 reformulated classification (RoI head, RoIAlign) (ablation)two-stage approach, ResNet-50 · LVIS-997-novelAP mask novel: 21.8 -> 22.8Improved
add stage I classification training (ablation)two-stage approach, ResNet-50 · LVIS-997-novelAP mask novel: 13.0 -> 22.8Improved
add reformulated classification (ablation)two-stage approach, ResNet-50 · ImageNet-1KTop-1 localization accuracy: 60.8 -> 64.7Improved
add stage I classification training (ablation)two-stage approach, ResNet-50 · ImageNet-1KTop-1 localization accuracy: 50.4 -> 64.7Improved
add reformulated classification (ablation)two-stage approach, ResNet-50 · ImageNet-1KGT-Known accuracy: 69.9 -> 73.8Improved
add stage I classification training (ablation)two-stage approach, ResNet-50 · ImageNet-1KGT-Known accuracy: 71.0 -> 73.8Improved
scale backbone from ResNet-50 to Swin-S (ablation)two-stage approach · ImageNet-1KTop-1 localization accuracy: 64.7 -> 68.3Improved
scale backbone from ResNet-50 to Swin-Stwo-stage approach · LVIS-997AP mask: 27.6 -> 35.1Improved
scale backbone from Swin-T to Swin-Stwo-stage approach · LVIS-997Delta AP (novel, all): -1.6 -> -4.5Hurt
use reverse flow (classifier first) instead of Deticdetector · ImageNet-21K benchmarkAP50: 1.3 -> 6.9Improved
successive IN-21K -> O365 training instead of classification-only pre-trainingLVIS fine-tuning · LVISAP: 35.8 -> 38.8Improved
joint training instead of successive training as pre-training (ablation)LVIS fine-tuning · LVISAP: 38.8 -> 40.1Improved

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