ybyangjing / cil-fce

Feature-Compatible Embeddings

This is the official PyTorch code for the paper:

A Class-Incremental Learning Approach for Learning Feature-Compatible Embeddings.

Hongchao An, Jing Yang*, Xiuhua Zhang, Yuankai Wu, Xiaoli Ruan, Shaobo Li, Jianjun Hu

This is the overall methodology diagram of the paper:

Fig. 1. FCE. More compatible embedding between the features of new and previous classes can be obtained by freezing the previous model and extending a new module.

Fig. 2. Fusion knowledge distillation. Removing the redundant feature dimensions resulting from extending the new modules in the first stage constrains the new and previous class features of the model to the plastic space and motivates the model to produce compatible embeddings between the features of the new and previous classes.

Prerequisites

The following packages are required to run the scripts:

• torch
• torchvision
• tqdm
• numpy

Training scripts

Train CIFAR-100: python main.py --config=./hyperparameter/b50inc10.json

Train ImageNet: python main.py --config=./hyperparameter/foster-imagenet100.json or python main.py --config=./hyperparameter/foster-imagenet1000.json

The complete code will be uploaded after the paper is accepted.