MNN: Mixed Nearest Neighbors for Self-Supervised Learning

Updata(12,Sep, 2024)

• MNN has been accepted in PR(Pattern Recognition)!!.

This is an PyTorch implementation of MNN proposed by our paper MNN: Mixed Nearest-Neighbors for Self-Supervised Learning. If you find this repo useful, welcome 🌟🌟🌟✨.

Installation

Step 0. Download and install Miniconda from official website

Step 1. Create a conda environment and activate it

conda create --name mnn python=3.9 -y
conda activate mnn

Step 2. Install PyTorch following official instructions, e.g.

pip install -r requirements.txt

Step 3. Install MNN

git clone https://github.com/pc-cp/MNN
cd MNN
chmod +x ./scripts.sh
./scripts.sh

Datasets:

• CIFAR10/CIFAR100: https://www.cs.toronto.edu/~kriz/cifar.html
• STL10: http://ai.stanford.edu/~acoates/stl10/stl10_binary.tar.gz
• Tiny ImageNet: http://cs231n.stanford.edu/tiny-imagenet-200.zip

The experimental environment in the manuscript

• Ubuntu 20.04.4 LTS (Focal Fossa)
• Python 3.9.7

Pre-trained Models

You can download pretrained models here:

• this link trained on three datasets.
• Download and place in the "./checkpoints" directory.

Results

Our model achieves the following performance:

Image Classification on four datasets

-	CIFAR-10	CIFAR-100	STL-10	Tiny ImageNet
MSF	89.94	59.94	88.05	42.68
MNN(Ours)	91.47	67.56	91.61	50.70

Citation

If you find this repo useful for your research, please consider citing the paper

@article{LONG2025110998,
title = {MNN: Mixed nearest-neighbors for self-supervised learning},
journal = {Pattern Recognition},
volume = {158},
pages = {110998},
year = {2025},
issn = {0031-3203},
doi = {https://doi.org/10.1016/j.patcog.2024.110998},
url = {https://www.sciencedirect.com/science/article/pii/S0031320324007490},
author = {Xianzhong Long and Chen Peng and Yun Li}

Contributors and Contact

📋 If there are any questions, feel free to contact with the authors.