AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Diffuse to Detect: Bi-Level Sample Rebalancing with Pseudo-Label Diffusion for Point-Supervised Infrared Small-Target Detection

Zhu Liu, Yuanhang Yao, Ping Qian, Zihang Chen, Risheng Liu

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 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 physics-induced diffusion labels to naive point supervision (ablation)ALCLNet single-point training · SIRST3IoU (%): 8.07 -> 69.64Improved
combine diffusion with superpixel prior (ablation)diffusion labels · SIRST3IoU (%): 69.64 -> 70.75Improved
add sample balance weights (meta-classifier) (ablation)full framework · SIRST3IoU (%): 70.96 -> 73.28Improved
add online label update (ablation)full framework · SIRST3IoU (%): 71.52 -> 73.28Improved
add dynamic aggregation (beta > 0) (ablation)full framework · SIRST3IoU (%): 67.74 -> 73.28Improved
use bi-level dual-update instead of direct joint learning (ablation)full framework · SIRST3IoU (%): 67.10 -> 73.28Improved
replace MCLC pseudo-label generation with physics-induced diffusionpseudo-mask generation · SIRST3 (centroid init)IoU (%): 66.98 -> 68.02Improved
replace MCLC pseudo-label generation with physics-induced diffusionpseudo-mask generation · SIRST-v1 (centroid init)IoU (%): 75.50 -> 74.58Hurt
replace MCLC pseudo-label generation with physics-induced diffusionpseudo-mask generation · per-target timeaverage time (s): 0.069s -> 0.012sImproved
train a detector from scratch on top 30% of samples ranked by learned weightsstandard detector · SIRST3IoUImproved
select samples by learned weights instead of random or hand-crafted difficulty (ablation)data selection · SIRST3IoUImproved
use diffusion instead of SAM2 for pseudo-maskspseudo-mask generation · SIRST3 (centroid init)IoU (%): 47.39 -> 68.02Improved

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