Awesome
[CVPR2023] Few-Shot Class-Incremental Learning via Class-Aware Bilateral Distillation
Requirements
Datasets
We follow FSCIL setting to use the same data index_list for training. Please follow the guidelines in CEC to prepare them. Scripts for experiments on mini-imagenet are as follows, and the full codes will be available upon acceptance:
Pretrain scripts
mini-imagenet (We also provide our pre-trained model so this step is optional.)
$ python train.py --dataset mini-imagenet --exp_dir experiment --epoch 200 --batch_size 256 --init_lr 0.1 --milestones 120 160 --val_start 100 --change_val_interval 160
Testing scripts
mini-imagenet
$ python test.py --dataset mini-imagenet --exp_dir experiment --needs_finetune --ft_iters 100 --ft_lr 0.001 --ft_factor 1.0 --ft_T 16 --w_d 100 --part_frozen --ft_KD_all --ft_teacher fixed --bilateral --BC_hidden_dim 64 --BC_lr 0.01 --w_BC_binary 50 --EMA_logits --w_l 1 --EMA_FC_lr 0.01
Acknowledgment
Our project references the codes in the following repos.