AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

From Infusion to Assimilation Distillation for Medical Image Segmentation

Jiankang Hong, Ye Luo, Yinan Liu, Junsong Yuan

CVPR 2026 · Label-efficient and robust learning › Knowledge distillation · approach: Distillation with causal and adaptive guidance

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 knowledge assimilation stage (KAS) after KIS (ablation)IAD · Synapseavg DICE: 83.13 -> 84.17Improved
add knowledge infusion stage (KIS) before KAS (ablation)IAD · Synapseavg DICE: 82.63 -> 84.17Improved
add KIS+KAS stages (ablation)IAD · ACDCavg DICE: 87.44 -> 89.29Improved
add class prototype loss L_cpd to soft-label loss (ablation)KIS · Synapseavg DICE: 82.18 -> 83.13Improved
add contrastive loss L_css to reverse feature constraint (ablation)KAS · Synapseavg DICE: 83.90 -> 84.17Improved
add L_css without L_rfc (ablation)KAS · Synapseavg DICE: 83.13 -> 82.48Hurt
use L2 rather than L1 or KLD distillation losses (ablation)IAD · Synapseavg DICE: 82.49 (L1), 81.87 (KLD) vs 84.17 (L2)Improved
add KAS plug-in to existing KD methods (ablation)OFD, CrossKD, CILD, DIST+, VL2Lite, SIKD · Synapseavg DICE: OFD 80.20 -> 83.24Improved
apply IAD to MobileNet-v2 and Tiny-ViT studentsIAD · Synapseavg DICE: 77.31 -> 82.06; 75.96 -> 81.26Improved
use class-weight ratio 1:2:4 for prototypes (ablation)KIS · SynapseDICE on rare classesImproved
add KAS to KIS for unseen transfer (ablation)IAD · STU (unseen)DICE: 64.01 -> 66.30Improved
add KIS-only distillation methods (e.g. KD, CATKD) vs no KD on transferKD baselines · ISIC2018/PH2/BUSI/STU transferavg DICE: 72.40 -> 68.78 (KD)Hurt

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