AI Paper Atlas

Research map / Detection and segmentation / CNN features

Explicit Visual Prompting for Universal Foreground Segmentations

Weihuang Liu, Xi Shen, Chi‐Man Pun, Xiaodong Cun

TPAMI 2025 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 25 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 frozen patch-embedding features Fpe as prompt (ablation)EVPv1 (SegFormer) · CASIA forgeryAUC: .833 -> .862Improved
add high-frequency features Fhfc as prompt (ablation)EVPv1 (SegFormer) · CASIA forgeryAUC: .846 -> .862Improved
add patch-embedding features Fpe as prompt (ablation)EVPv2 (SegFormer) · ISTD shadowBER: 1.75 -> 1.35Improved
add Frequency Enhanced Adaptor instead of two plain MLP blocks (ablation)EVPv2 (SegFormer) · ISTD shadowBER: 1.40 -> 1.35Improved
unshare MLP_tune across adaptors (shared MLP_tune) (ablation)EVPv1 (SegFormer) · ISTD shadowBER: 1.77 -> 1.35Improved
use unshared MLP_up in each adaptor (ablation)EVPv1 (SegFormer) · ISTD shadowBER: 1.35 -> 1.33No clear effect
tune all four SegFormer stages instead of only Stage1 (ablation)EVPv2 (SegFormer) · ISTD shadowBER: 3.99 -> 1.35Improved
tune Stage3 only vs all stages (ablation)EVPv2 (SegFormer) · ISTD shadowBER: 1.44 -> 1.35Improved
reduce scale factor r from 64 to 16 (more tunable parameters) (ablation)EVPv2 (SegFormer) · ISTD shadowBER: 1.69 -> 1.35Improved
reduce scale factor r from 16 to 4 (ablation)EVPv2 (SegFormer) · ISTD shadowBER: 1.35 -> 1.53Hurt
replace full fine-tuning with EVPv2 (frozen backbone) (ablation)SegFormer · ISTD shadowBER: 2.42 -> 1.35Improved
replace full fine-tuning with EVPv2 (frozen backbone) (ablation)ViT-CLS · CASIA forgeryAUC: .692 -> .784Improved

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