Research map / Continual learning / Continual fine-tuning of LLMs
SAMCL: Empowering SAM to Continually Learn from Dynamic Domains with Extreme Storage Efficiency
Zeqing Wang, Kangye Ji, Di Wang, Haibin Zhang, Fei Cheng
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add AugModule (SLoRA + prompt augmentation) to LoRA baseline (ablation) | SAMCL (SAM) · CL Kvasir to COD | AA mIoU: 0.703 -> 0.748 | Improved |
| add Module Selector on top of AugModule (ablation) | SAMCL (SAM) · CL Kvasir to COD | AA mIoU: 0.748 -> 0.836 | Improved |
| add Module Selector on top of AugModule (ablation) | SAMCL (SAM) · CL Kvasir to COD | FM mIoU: 0.086 -> 0.0019 | Improved |
| add Module Selector on top of AugModule (ablation) | SAMCL (SAM) · CL Kvasir to COD | FT mIoU: 0.570 -> 0.668 | Improved |
| add prompt augmentation (AugModule instead of SLoRA) (ablation) | SLoRA (SAM) · COD | mIoU: 0.735 -> 0.782 | Improved |
| share matrix A across layers (SLoRA instead of LoRA) (ablation) | LoRA (SAM) · COD | mIoU: 0.733 -> 0.735 | No clear effect |
| share matrix A across layers (SLoRA instead of LoRA) (ablation) | LoRA (SAM) · Parameter storage | MB: 0.308 -> 0.190 | Improved |
| store 300 embeddings per task instead of 10 (ablation) | SAMCL (SAM) · CL Kvasir to COD | FM mIoU: -0.0546 -> 0.0009 | Improved |
| store 500 embeddings per task instead of 10 (ablation) | SAMCL (SAM) · CL Kvasir to COD | AA mIoU: 0.808 -> 0.838 | Improved |
| train Module Selector on COCO subset after CL (ablation) | SAMCL (SAM) · Revert to original SAM | mIoU: 0.5281 -> 0.6162 | Improved |
| use SAMCL instead of MoDA | MoDA (SAM) · CL Kvasir to COD | AA mIoU: 0.756 -> 0.836 | Improved |
| use SAMCL instead of ER (replay) | ER (SAM) · CL Kvasir to COD | FT mIoU: 0.630 -> 0.668 | Improved |
Closest papers by meaning
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.87)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.87)
- TreeLoRA: Efficient Continual Learning via Layer-Wise LoRAs Guided by a Hierarchical Gradient-Similarity Tree (ICML 2025 · similarity 0.86)
- SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (ICLR 2025 · similarity 0.85)
- Mutual Learning for SAM Adaptation: A Dual Collaborative Network Framework for Source-Free Domain Transfer (ICML 2025 · similarity 0.85)
- Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (NeurIPS 2024 · similarity 0.85)
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.
