Continuous Knowledge-Preserving Decomposition with Adaptive Layer Selection for Few-Shot Class-Incremental Learning

This is the official repository for Continuous Knowledge-Preserving Decomposition with Adaptive Layer Selection for Few-Shot Class-Incremental Learning, accepted at ACM Multimedia 2026 (ACM MM 2026).

Continuous Knowledge-Preserving Decomposition with Adaptive Layer Selection for Few-Shot Class-Incremental Learning Xiaojie Li, Wei Liu, Bei Wang, Jianlong Wu, Yue Yu, Min Zhang ACM Multimedia 2026

🔨 Installation

1. Create a Conda environment:

   conda create --name ckpdfscil python=3.10 -y
   conda activate ckpdfscil

2. Install dependencies:

   pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 torchaudio==0.12.1 \
     --extra-index-url https://download.pytorch.org/whl/cu113
   
   pip install -U openmim
   mim install mmcv-full==1.7.0 mmengine==0.10.4
   
   pip install opencv-python matplotlib einops timm==0.6.12 \
     scikit-learn transformers==4.44.2
   
   pip install git+https://github.com/openai/CLIP.git
   
   git clone https://github.com/state-spaces/mamba.git
   cd mamba
   git checkout v1.2.0.post1
   pip install .
   cd ..

3. Clone this repository:

   git clone https://github.com/xiaojieli0903/CKPD-FSCIL.git
   cd CKPD-FSCIL
   mkdir -p ./data

---

➡️ Data Preparation

1. Download the datasets from the NC-FSCIL dataset repository.

2. Organize the datasets as follows:

   ./data/
   ├── cifar/
   ├── CUB_200_2011/
   └── miniimagenet/
   

---

➡️ Pretrained Model Preparation

Use tools/convert_pretrained_model.py to convert pretrained models into the required format.

The following model types are supported:

• CLIP: converts OpenAI CLIP models.
• TIMM: converts TIMM models.

CLIP Model

python tools/convert_pretrained_model.py \
  ViT-B/32 \
  ./pretrained_models/clip-vit-base-p32_openai.pth \
  --model-type clip

TIMM Model

python tools/convert_pretrained_model.py \
  vit_base_patch16_224 \
  ./pretrained_models/vit_base_patch16_224.pth \
  --model-type timm

---

🚀 Training

Execute the corresponding script to start training.

MiniImageNet

sh train_miniimagenet.sh

CUB-200-2011

sh train_cub.sh

---

✏️ Citation

If you find this work useful in your research, please consider citing:

@inproceedings{li2026continuous,
  title     = {Continuous Knowledge-Preserving Decomposition with Adaptive Layer Selection for Few-Shot Class-Incremental Learning},
  author    = {Li, Xiaojie and Liu, Wei and Wang, Bei and Wu, Jianlong and Yu, Yue and Zhang, Min},
  booktitle = {Proceedings of the ACM International Conference on Multimedia},
  year      = {2026}
}