AI Paper Atlas

Research map / Detection and segmentation / CNN features

Feature Pyramid Transformer

Dong Zhang, Hanwang Zhang, Jinhui Tang, Meng Wang, Xiansheng Hua, Qianru Sun

ECCV 2020 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 323 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 self-transformer (ST) (ablation)BFP · COCO 2017 val (detection/instance seg)box AP: 31.6 -> 32.0Improved
add grounding transformer (GT) (ablation)BFP · COCO 2017 val (detection/instance seg)box AP: 31.6 -> 35.1Improved
add rendering transformer (RT) (ablation)BFP · COCO 2017 val (detection/instance seg)box AP: 31.6 -> 34.7Improved
add all three transformers (full FPT) (ablation)BFP · COCO 2017 val (detection/instance seg)box AP: 31.6 -> 38.0Improved
add SBN to FPT (ablation)FPT · COCO 2017 val (detection/instance seg)box AP: 37.2 -> 37.8Improved
add DropBlock to FPT (ablation)FPT · COCO 2017 val (detection/instance seg)box AP: 37.2 -> 37.5Improved
add SBN and DropBlock together to FPT (ablation)FPT · COCO 2017 val (detection/instance seg)box AP: 37.2 -> 38.0Improved
use Euclidean similarity instead of softmax dot-product in GT (ablation)BFP+GT · COCO 2017 val (detection/instance seg)box AP: 32.2 -> 33.7Improved
add mixture of softmaxes (MoS) to Euclidean GT (ablation)BFP+GT · COCO 2017 val (detection/instance seg)box AP: 33.7 -> 34.7Improved
increase MoS components N in ST from 2 to 8 (ablation)BFP+ST · COCO 2017 val (detection/instance seg)box AP: 31.8 -> 30.0Hurt
increase MoS components N in GT from 4 to 8 (ablation)BFP+GT · COCO 2017 val (detection/instance seg)box AP: 34.7 -> 32.5Hurt
enlarge LGT square_size from 5 to 9 (ablation)UFP+LGT · Cityscapes valmIoU: 80.0 -> 79.6Hurt

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