/Fine-Grained-Continual-Learning

Implementation of AR1* and CWR* Continual Learning techniques for the NICv2 scenario

Primary LanguagePythonBSD 3-Clause Clear LicenseBSD-3-Clause-Clear

Rehearsal-Free Continual Learning over Small Non-I.I.D. Batches

This is the original Caffe implementation of AR1* and CWR* Continual Learning techniques.

A custom Caffe distribution packaged as a Docker image is used. More info and source code can be found here. The Docker image is already available on the Docker Hub here.

See the Running the experiments section for a detailed guide on how to reproduce our experiments.

An official PyTorch implementation of the AR1* and CWR* algorithms is also available here!

Reference

Our article "Rehearsal-Free Continual Learning over Small Non-I.I.D. Batches" is now available here!

@InProceedings{lomonaco2019nicv2,
   title = {Rehearsal-Free Continual Learning over Small Non-I.I.D. Batches},
   author = {Vincenzo Lomonaco and Davide Maltoni and Lorenzo Pellegrini},
   journal = {1st Workshop on Continual Learning in Computer Vision at CVPR2020},
   url = "https://arxiv.org/abs/1907.03799",
   year = {2019}
}

Running the experiments

Some helper scripts are provided under the Run experiments folder.

You can run an experiment as follows.

  1. Install the Nvidia Docker Toolkit from here

  2. Move inside the Run experiments folder:

cd "Run experiments"
  1. Prepare the project source and create the bash script. This can be achieved by issuing the following command:
python prepare_experiment.py method scenario path-to-core50 [--nvidia_docker x]

where method can be "CWR", "AR1" or "Naive" and scenario can be "79", "196" or "391". You can also execute the script with a single argument "-h" to view a description of the expected parameters.

You can set the desidered Nvidia Docker run method by passing either:

  • "--nvidia_docker nvidia-docker2" more info here
  • "--nvidia_docker nvidia-container-toolkit" more info here as an argument. Defaults to nvidia-docker2.

When passing the "path-to-core50" argument, make sure that the selected folder contains the following content:

  • A file named core50_labels.txt, containing the Core50 labels. Can be downloaded here
  • A folder named core50_128x128 containing the 128x128 version of the CORe50 dataset. Can be downloaded here

  1. Running the python script will have the following effects:

    • Copy the correct prototxt files inside the Project Source/NIC_v2/NIC_v2_X folder
    • Create a proper exp_configuration.json" file inside the Project Source folder
    • Create a run_experiment.sh file inside the Run experiments folder. Should be already executable when created

  2. Execute the run_experiment.sh script as follows:

./run_experiment.sh

This will run the experiment on "run0" inside our docker image in interactive mode (issuing CTRL+C or closing the terminal will terminate the experiment).

Content

The content can be summarized as follows:

  • The project source code (Project Source folder)
    • inc_training_Core50.py contains the entry point;
    • nicv2_configuration.py contains the experiment configuration loader;
    • The filelists used (batch_filelists subdirectory)
      • As explained in our paper, we used a reduced version of the original Core50 test set (test_filelist_20.txt);
      • For the proposed scenarios (NICv2 79, 196, 391) we provide a separate folder, each containing a sub-directory for each run. We used all the provided 10 runs in order to obtain the average test accuracy curves reported in our paper;
    • A MobileNetV1 pretrained with ImageNet (models/MobileNetV1.caffemodel)
    • The prototxt(s) describing the solvers and nets (NIC_v2 folder)
  • A set of configuration scripts to run the experiments (Run experiments folder)

Core50 Dataset

The Core50 Dataset can be downloaded from https://vlomonaco.github.io/core50/index.html#download In our test we used the 128x128 version, zip archive.