Inverse Design of Photonic Crystal Lasers Through Automatic Differentiation Optimization

Mihai Crisan

doi:10.1109/jstqe.2026.3687110

AF-CANet: Augmentation-Free, Curriculum-Guided Attention UNet Framework for Few-shot Document Layout Analysis

2026-05-09

0

PRE

AI

H

Hadia Showkat Kawoosa *

S

Sahana Rangasrinivasan

S

Srirangaraj Setlur

P

Puneet Goyal

V

Venu Govindaraju

DOI：10.1016/j.patrec.2026.05.003

Original

Original request for help

Save

Abstract

En 中文

• A novel framework, AF-CANet, for few-shot, many-shot and binarized DLA. • Efficient Dilated Channel Attention extracts rich multiscale semantic features. • Cross-dataset curriculum learning improves generalization across manuscripts. • Weighted Dice loss mitigates class imbalance in segmentation tasks. • Experiments on three benchmarks show consistent gains over state-of-the-art methods.

Keywords:

Few-shot learning

Document layout analysis

Attention mechanism

Curriculum learning

Segmentation

AF-CANet: Augmentation-Free, Curriculum-Guided Attention UNet Framework for Few-shot Document Layout Analysis

Abstract

Journal

Organization

Cited Papers

Citing Papers