Implementation for paper "Towards the Generalization of Contrastive Self-Supervised Learning"

Last update: Nov 30, 2022

Overview

Contrastive Self-Supervised Learning on CIFAR-10

Paper

"Towards the Generalization of Contrastive Self-Supervised Learning", Weiran Huang, Mingyang Yi and Xuyang Zhao, arXiv:2111.00743, 2021.

The most critical argument we made in our paper is that the quality of data augmentation exhibits great impact on the quality of contrastive-learned encoder. The data augmentation with sharper intra-class concentration enables the model to have better generalization on downstream tasks. We verify it through a variety of experiments in this repository.

Supported methods

SimCLR
Barlow Twins
MoCo
SimSiam

Installation

pip install -r requirement.txt

Dependencies

torch==1.4.0
torchvision==0.5.0
pytorch-lightning==1.3.8
lightly==1.0.8 (important!)

Evaluation

KNN evaluation protocol. Code from here.

Results

ResNet-18 trained by SimCLR and Barlow Twins over various data augmentation combinations.

Example: python main.py --model=twins --epoch=800 --batch=512 --round=3 --augs=abcde

(a)	(b)	(c)	(d)	(e)	SimCLR	Barlow Twins
✓	✓	✓	✓	✓	89.92 ± 0.05	83.93 ± 0.57
✓	✓	✓	✓	×	88.41 ± 0.11	83.37 ± 0.43
✓	✓	✓	×	×	83.62 ± 0.19	73.70 ± 0.99
✓	✓	×	×	×	62.91 ± 0.25	49.56 ± 0.11
✓	×	×	×	×	62.37 ± 0.09	48.54 ± 0.29

Augmentation operations include:

(a) random cropping with a scaling factor chosen in [0.08, 1.0];

(b) random Gaussian blur with a probability 0.5;

(d) color distortion with a probability of 0.8 and with strength of [0.4, 0.4, 0.4, 0.1];

(e) random horizontal flipping with a probability of 0.5.

ResNet18 trained by SimCLR and Barlow Twins over various color distortion strengths.

Example: python main.py --model=simclr --epoch=800 --batch=512 --round=3 --augs=color --strength=1

Color Distortion Strength	SimCLR	Barlow Twins
1/8	73.60 ± 0.11	61.13 ± 2.81
1/4	76.25 ± 0.16	68.30 ± 0.15
1/2	78.49 ± 0.09	72.76 ± 1.50
1	82.64 ± 0.57	78.79 ± 0.54

Acknowledgement

This code is based on:

Citation

If you find our work useful in your research, please consider citing:

@misc{huang2021generalization,
      title={Towards the Generalization of Contrastive Self-Supervised Learning}, 
      author={Weiran Huang and Mingyang Yi and Xuyang Zhao},
      year={2021},
      eprint={2111.00743},
      archivePrefix={arXiv}
}

You might also like...

Implementation of Nyström Self-attention, from the paper Nyströmformer

Nyström Attention Implementation of Nyström Self-attention, from the paper Nyströmformer. Yannic Kilcher video Install $ pip install nystrom-attention

95 Jan 2, 2023

Implementation of SETR model, Original paper: Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.

SETR - Pytorch Since the original paper (Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.) has no official

112 Dec 16, 2022

Official implementation of the paper Image Generators with Conditionally-Independent Pixel Synthesis https://arxiv.org/abs/2011.13775

CIPS -- Official Pytorch Implementation of the paper Image Generators with Conditionally-Independent Pixel Synthesis Requirements pip install -r requi

Multimodal Lab @ Samsung AI Center Moscow

201 Dec 21, 2022

Official pytorch implementation of paper

Implementation for paper "Towards the Generalization of Contrastive Self-Supervised Learning"

Related tags

Overview

Contrastive Self-Supervised Learning on CIFAR-10

Paper

Supported methods

Installation

Dependencies

Evaluation

Results

ResNet-18 trained by SimCLR and Barlow Twins over various data augmentation combinations.

ResNet18 trained by SimCLR and Barlow Twins over various color distortion strengths.

Acknowledgement

Citation

You might also like...

Implementation of Nyström Self-attention, from the paper Nyströmformer

Implementation of SETR model, Original paper: Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers.

Official implementation of the paper Image Generators with Conditionally-Independent Pixel Synthesis https://arxiv.org/abs/2011.13775

Official pytorch implementation of paper "Image-to-image Translation via Hierarchical Style Disentanglement".

PyTorch implementation of paper "Neural Scene Flow Fields for Space-Time View Synthesis of Dynamic Scenes", CVPR 2021

Implementation of Barlow Twins paper

Official pytorch implementation of paper "Inception Convolution with Efficient Dilation Search" (CVPR 2021 Oral).

Official implementation of our paper "LLA: Loss-aware Label Assignment for Dense Pedestrian Detection" in Pytorch.

Functional TensorFlow Implementation of Singular Value Decomposition for paper Fast Graph Learning

Releases(v0.2.1)

v0.2.1(Jun 16, 2022)

v0.2.0(Jun 4, 2022)

v0.1.0(Jun 4, 2022)

Owner

Weiran Huang

DeLag: Detecting Latency Degradation Patterns in Service-based Systems

Anime Face Detector using mmdet and mmpose

Code and dataset for ACL2018 paper "Exploiting Document Knowledge for Aspect-level Sentiment Classification"

Computational Methods Course at UdeA. Forked and size reduced from:

Codes for AAAI 2022 paper: Context-aware Health Event Prediction via Transition Functions on Dynamic Disease Graphs

Code for paper Novel View Synthesis via Depth-guided Skip Connections

Code for: Imagine by Reasoning: A Reasoning-Based Implicit Semantic Data Augmentation for Long-Tailed Classification

Symbolic Parallel Adaptive Importance Sampling for Probabilistic Program Analysis in JAX

PyTorch Implementations for DeeplabV3 and PSPNet

DECAF: Deep Extreme Classification with Label Features

A simplistic and efficient pure-python neural network library from Phys Whiz with CPU and GPU support.

A Python library that provides a simplified alternative to DBAPI 2

Birthday-problem - The birthday problem asks for the probability that, in a set of n randomly chosen people, at least two will share a birthday

Code for "Single-view robot pose and joint angle estimation via render & compare", CVPR 2021 (Oral).

Deep Learning for Natural Language Processing SS 2021 (TU Darmstadt)

K Closest Points and Maximum Clique Pruning for Efficient and Effective 3D Laser Scan Matching (To appear in RA-L 2022)

Elucidating Robust Learning with Uncertainty-Aware Corruption Pattern Estimation

This is a simple plugin for Vim that allows you to use OpenAI Codex.

AFL binary instrumentation

Code for reproducing our analysis in the paper titled: Image Cropping on Twitter: Fairness Metrics, their Limitations, and the Importance of Representation, Design, and Agency