Learning Optimal Inter-class Margin Adaptively for Few-Shot Class-Incremental Learning via Neural Collapse-based Meta-learning

Hang Ran, Weijun Li, Lusi Li, Songsong Tian, Xin Ning, Prayag Tiwari

Requirements

The conda software is required for running the code. Generate a new environment with

$ conda create --name metanc_env python=3.6
$ conda activate metanc_env

We need PyTorch 1.3 and CUDA.

$ (metanc_env) conda install pytorch=1.3 torchvision cudatoolkit=10.1 -c pytorch
$ (metanc_env) pip install -r requirements.txt

Datasets

We provide the code for running experiments on miniImageNet, CIFAR100 and CUB200. For CIFAR100, the dataset will be download automatically. For miniImageNet, you can download the dataset here. Please put the downloaded file under src/data/ folder and unzip it.

$ (metanc_env) cd src/data/
$ (metanc_env) gdown 1_x4o0iFetEv-T3PeIxdSbBPUG4hFfT8U
$ (metanc_env) tar -xvf miniimagenet.tar 

Usage

The whole simulator is runnable from the command line via the src/main.py script which serves as a command parser. Everything should be run from the src directory.

The structure of any command looks like

$ (metanc_env) python main.py command [subcommand(s)] [-option(s)] [argument(s)]

The main.py file also contains all default parameters used for simulations.

Simulation

To run a single simulation of the model (incl. training, validation, testing), use the simulation command. A logging directory should be specified, in case the default path is not wanted. Any simulation parameter that should be different from the default found in main.py can be specified by chaining -p parameter value pairs.

$ (metanc_env) python main.py simulation --logdir path/to/logdir -p parameter_1 value_1 -p parameter_2 value_2

All parameters are interpreted as strings and translated by the parser, so no "s are needed. Boolean parameters' value can be specified as t, true, f or false.

Run main experiments on CIFAR100

# Pretraining
$ (metanc_env) python -u main.py simulation -v -ld "log/test_cifar100/pretrain_basetrain" -p max_train_iter 120 -p data_folder "data" -p trainstage pretrain_baseFSCIL -p dataset cifar100 -p learning_rate 0.01 -p batch_size 32 -p optimizer SGD -p SGDnesterov True -p lr_step_size 30 -p dim_features 512 -p block_architecture mini_resnet12 -p num_workers 8
# Metatraining
$ (metanc_env) python -u main.py simulation -v -ld "log/test_cifar100/meta_basetrain" -p max_train_iter 70000 -p data_folder "data" -p resume "log/test_cifar100/pretrain_basetrain"  -p trainstage metatrain_baseFSCIL -p dataset cifar100 -p average_support_vector_inference True -p learning_rate 0.01 -p batch_size_training 10 -p batch_size_inference 128 -p optimizer SGD -p SGDnesterov True -p lr_step_size 30000 -p dim_features 512 -p num_ways 60 -p num_shots 5 -p block_architecture mini_resnet12
# Evaluation(num_shots relates only to number of shots in base session, on novel there are always 5)
$ (metanc_env) python -u main.py simulation -v -ld "log/test_cifar100/eval/mode1"  -p data_folder "data"  -p resume "log/test_cifar100/meta_basetrain" -p dim_features 512  -p trainstage train_FSCIL -p dataset cifar100 -p learning_rate 0.01 -p batch_size_training 128 -p batch_size_inference 128 -p num_query_training 0 -p optimizer SGD -p SGDnesterov True -p retrain_act tanh -p num_ways 60 -p num_shots 200 -p block_architecture mini_resnet12

Run main experiments on miniImageNet

# Pretraining
$ (metanc_env) python -u main.py simulation -v -ld "log/test_miniImageNet/pretrain_basetrain" -p max_train_iter 120 -p data_folder "data" -p trainstage pretrain_baseFSCIL -p dataset miniImageNet -p learning_rate 0.01 -p batch_size 32 -p optimizer SGD -p SGDnesterov True -p lr_step_size 30 -p dim_features 512 -p block_architecture mini_resnet12 -p num_workers 8
# Metatraining
$ (metanc_env) python -u main.py simulation -v -ld "log/test_miniImageNet/meta_basetrain" -p max_train_iter 70000 -p data_folder "data" -p resume "log/test_miniImageNet/pretrain_basetrain"  -p trainstage metatrain_baseFSCIL -p dataset miniImageNet -p average_support_vector_inference True -p learning_rate 0.01 -p batch_size_training 10 -p batch_size_inference 128 -p optimizer SGD -p SGDnesterov True -p lr_step_size 30000 -p dim_features 512 -p num_ways 60 -p num_shots 5 -p block_architecture mini_resnet12
# Evaluation(num_shots relates only to number of shots in base session, on novel there are always 5)
$ (metanc_env) python -u main.py simulation -v -ld "log/test_miniImageNet/eval/mode1"  -p data_folder "data"  -p resume "log/test_miniImageNet/meta_asetrain" -p dim_features 512  -p trainstage train_FSCIL -p dataset miniImageNet -p learning_rate 0.01 -p batch_size_training 128 -p batch_size_inference 128 -p num_query_training 0 -p optimizer SGD -p SGDnesterov True -p retrain_act tanh -p num_ways 60 -p num_shots 200 -p block_architecture mini_resnet12

Run main experiments on cub200

# Pretraining
$ (metanc_env) python -u main.py simulation -v -ld "log/test_cub200/pretrain_basetrain" -p max_train_iter 120 -p data_folder "data" -p trainstage pretrain_baseFSCIL -p dataset miniImageNet -p learning_rate 0.01 -p batch_size 32 -p optimizer SGD -p SGDnesterov True -p lr_step_size 30 -p dim_features 512 -p block_architecture mini_resnet18 -p num_workers 8
# Metatraining
$ (metanc_env) python -u main.py simulation -v -ld "log/test_cub200/meta_basetrain" -p max_train_iter 70000 -p data_folder "data" -p resume "log/test_cub200/pretrain_basetrain"  -p trainstage metatrain_baseFSCIL -p dataset cub200 -p average_support_vector_inference True -p learning_rate 0.01 -p batch_size_training 10 -p batch_size_inference 128 -p optimizer SGD -p SGDnesterov True -p lr_step_size 30000 -p dim_features 512 -p num_ways 60 -p num_shots 5 -p block_architecture mini_resnet18
# Evaluation(num_shots relates only to number of shots in base session, on novel there are always 5)
$ (metanc_env) python -u main.py simulation -v -ld "log/test_cub200/eval/mode1"  -p data_folder "data"  -p resume "log/test_cub200/meta_asetrain" -p dim_features 512  -p trainstage train_FSCIL -p dataset cub200 -p learning_rate 0.01 -p batch_size_training 128 -p batch_size_inference 128 -p num_query_training 0 -p optimizer SGD -p SGDnesterov True -p retrain_act tanh -p num_ways 60 -p num_shots 200 -p block_architecture mini_resnet18

Inspection with TensorBoard

For a detailed inspection of the simulation, the TensorBoard tool can be used. During simulations, data is collected which can be illustrated by the tool in the browser.

Acknowledgment

Our code is based on

• CFSCIL (Main framework)
• ALICE(Data augmentation)