AI Paper Atlas

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

IP-SAM: Rethinking Prompt-Conditioned Segmentation for Prompt-Absent Deployment

Huiyao Zhang, Jin Bai, Rui Guo, Jianwen Tan, Hongfei Wang, Ye Li

ECCV 2026 · Detection and segmentation › Open-set and few-shot detection · approach: Interactive and referring segmentation · 0 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-Prompt Generator (SPG) with prompt-space conditioning (ablation)IP-SAM baseline · COD10KMAE: 0.024 -> 0.022Improved
add Prompt-Space Gating (PSG) after SPG (ablation)IP-SAM · CAMOMAE: 0.041 -> 0.034Improved
add refinement head to full model without it (ablation)IP-SAM · COD10KMAE: 0.018 -> 0.017Improved
route SPG output through frozen prompt encoder and decoder instead of using it directly as mask (ablation)IP-SAM · COD10KMAE: 0.025 -> 0.017Improved
prompt-space conditioning with low-capacity Minimal-SPG (ablation)Minimal-SPG · COD10KMAE: 0.051 -> 0.040Improved
condition through prompt space instead of adding dense prompts to image features (FeatAdd) (ablation)IP-SAM · COD10KMAE: 0.020 -> 0.017Improved
keep prompt encoder frozen instead of trainable (ablation)IP-SAM · COD10KMAE: 0.019 -> 0.017Improved
use asymmetric gating instead of subtraction fusion of Z+ and Z- (ablation)PSG · COD10KMAE: 0.019 -> 0.017Improved
anchor PSG residual to original Z+ instead of suppressed Z+ (ablation)PSG · COD10KMAE: 0.019 -> 0.017Improved
use propagated SPG output tokens instead of default SAM2 mask tokens (ablation)IP-SAM · COD10KMAE: 0.019 -> 0.017Improved
apply LoRA to all blocks instead of deep stages only (ablation)IP-SAM · COD10KMAE: 0.034 -> 0.017Improved
scale backbone down from SAM2-L to SAM2-T (ablation)IP-SAM · COD10KMAE: 0.017 -> 0.080Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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