This repository provides a blank continual/incremental learning template. The template does not add any continual/incremental learning strategy and uses finetuning for incremental learning.
This means that the approach will suffer catastrophic forgetting.
You can use this blank template to reduce catastrophic forgetting by adding your own strategies!
We hope this template helps people new to incremental learning to better understand the process.
-
We provide some registration methods for computer vision data, remote sensing image data
-
We provide some tools to measure the computational efficiency of the model, etc.
-
We design the interface of some classic datasets and backbone network, so that you can make your own model modification and comparison more conveniently
We will provide a repository of reproduction code for classic work based on this template. Any suggestions and questions are welcome~