AI Paper Atlas

Research map / Continual learning / Class-incremental

Semantic-Guided Global-Local Collaborative Prompt Learning for Few-Shot Class Incremental Learning

Yongxin Yan, Weisen Chen, Xingye Chen, Yuanjie Shao, Zhengrong Zuo, Wenming Tan, Wenqi Ren, Changxin Gao, Nong Sang

CVPR 2026 · Continual learning › Class-incremental · approach: Few-shot class-incremental prototypes

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 Global-Level prompt learning to the CLIP baseline (ablation)SGLC · CUB200 (FSCIL)Aavg: 58.09 -> 78.04Improved
add Local-Level prompt learning (optimal transport) on top of global (ablation)SGLC · CUB200 (FSCIL)Aavg: 78.04 -> 79.08Improved
add Semantic-Guided LLM description distillation (ablation)SGLC · CUB200 (FSCIL)Aavg: 79.08 -> 80.18Improved
add learnable deep-layer prompts to CoOp (ablation)CoOp · CUB200 (FSCIL)Aavg: 69.47 -> 71.22Improved
add adapter-based visual feature correction to CoOp (ablation)CoOp · CUB200 (FSCIL)Aavg: 69.47 -> 70.72Improved
combine deep-layer prompts and visual feature correction (ablation)CoOp · CUB200 (FSCIL)Aavg: 69.47 -> 72.53Improved
use more LLM-selected attributes (add pattern) (ablation)SGLC local prompts · CUB200 (FSCIL)Aavg: 80.18 -> 79.63Hurt
use four attributes (add pattern and size) (ablation)SGLC local prompts · CUB200 (FSCIL)Aavg: 80.18 -> 79.01Hurt
replace one-to-one cosine alignment with optimal transport in local alignment (ablation)SGLC · CUB200 (FSCIL)accuracy per sessionImproved
use category-specific prompts with frozen old-class prompts instead of category-agnostic prompts (ablation)SGLC global prompts · CUB200 (FSCIL)base-class accuracy across sessionsImproved
train only 0.71M parameters via prompts and adapter instead of full fine-tuningSGLC · CUB200 (FSCIL)trainable parameters: 86.68M -> 0.71MImproved

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