Official PyTorch implementation of the paper:

All4One: Symbiotic Neighbour Contrastive Learning via Self-Attention and Redundancy Reduction

Imanol G. Estepa, Ignacio Sarasúa, Bhalaji Nagarajan, and Petia Radeva

To enhance the fair comparison of different self-supervised methods, the code is provided as an addition to solo-learn SSL library. Even so, trained models can be used as backbones for tasks not included in this repository.

First clone the repo.

Then, to install solo-learn with Dali and/or UMAP support, use:

pip3 install .[dali,umap,h5] --extra-index-url https://developer.download.nvidia.com/compute/redist

If no Dali/UMAP/H5 support is needed, the repository can be installed as:

pip3 install .

Additionally, All4One requires the installation of positional-encodings dependency, it can be installed with the following command:

pip3 install positional-encodings

The exact dependency versions can be found on the requirements.txt file

Training configurations are provided in scripts/pretrain/ folder. They can be executed by:

python3 main_pretrain.py \
    # path to training script folder
    --config-path scripts/pretrain/cifar/ \
    # training config name
    --config-name all4one.yaml
    # add new arguments (e.g. those not defined in the yaml files)
    # by doing ++new_argument=VALUE
    # pytorch lightning's arguments can be added here as well.

launcher.py file wraps up the command to initialize the training with:

python3 launcher.py

For downstream task evaluation, we recommend following official solo-learn tutorials.

🚀 methods where hyperparameters were heavily tuned. More results and ablations in the paper.

If you find All4One useful, please cite our paper:

@InProceedings{Estepa_2023_ICCV,
    author    = {Estepa, Imanol G. and Sarasua, Ignacio and Nagarajan, Bhalaji and Radeva, Petia},
    title     = {All4One: Symbiotic Neighbour Contrastive Learning via Self-Attention and Redundancy Reduction},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {16243-16253}
}

This work was partially funded by the Horizon EU project MUSAE (No. 01070421), 2021-SGR-01094 (AGAUR), Icrea Academia’2022 (Generalitat de Catalunya), Robo STEAM (2022-1-BG01-KA220-VET-000089434, Erasmus+ EU), DeepSense(ACE053/22/000029, ACCIO), DeepFoodVol (AEI-MICINN, PDC2022-133642-I00), PID2022-141566NB-I00 (AEI-MICINN), and CERCA Programme / Generalitatde Catalunya. B. Nagarajan acknowledges the support of FPI Becas, MICINN, Spain.