https://github.com/sthalles/simclr

PyTorch implementation of SimCLR: A Simple Framework for Contrastive Learning of Visual Representations
https://github.com/sthalles/simclr

contrastive-loss deep-learning machine-learning pytorch pytorch-implementation representation-learning simclr torchvision unsupervised-learning

Last synced: 29 days ago
JSON representation

PyTorch implementation of SimCLR: A Simple Framework for Contrastive Learning of Visual Representations

• Host: GitHub
• URL: https://github.com/sthalles/simclr
• Owner: sthalles
• License: mit
• Created: 2020-02-17T18:58:35.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2024-03-04T10:45:52.000Z (8 months ago)
• Last Synced: 2024-10-15T00:23:44.061Z (29 days ago)
• Topics: contrastive-loss, deep-learning, machine-learning, pytorch, pytorch-implementation, representation-learning, simclr, torchvision, unsupervised-learning
• Language: Jupyter Notebook
• Homepage: https://sthalles.github.io/simple-self-supervised-learning/
• Size: 80.5 MB
• Stars: 2,240
• Watchers: 20
• Forks: 464
• Open Issues: 30
• Metadata Files:
  • Readme: README.md
  • License: LICENSE.txt

Awesome Lists containing this project

README

        
# PyTorch SimCLR: A Simple Framework for Contrastive Learning of Visual Representations

[![DOI](https://zenodo.org/badge/241184407.svg)](https://zenodo.org/badge/latestdoi/241184407)

### Blog post with full documentation: [Exploring SimCLR: A Simple Framework for Contrastive Learning of Visual Representations](https://sthalles.github.io/simple-self-supervised-learning/)

![Image of SimCLR Arch](https://sthalles.github.io/assets/contrastive-self-supervised/cover.png)

### See also [PyTorch Implementation for BYOL - Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning](https://github.com/sthalles/PyTorch-BYOL).

## Installation

```

$ conda env create --name simclr --file env.yml

$ conda activate simclr

$ python run.py

```

## Config file

Before running SimCLR, make sure you choose the correct running configurations. You can change the running configurations by passing keyword arguments to the ```run.py``` file.

```python

$ python run.py -data ./datasets --dataset-name stl10 --log-every-n-steps 100 --epochs 100 

```

If you want to run it on CPU (for debugging purposes) use the ```--disable-cuda``` option.

For 16-bit precision GPU training, there **NO** need to to install [NVIDIA apex](https://github.com/NVIDIA/apex). Just use the ```--fp16_precision``` flag and this implementation will use [Pytorch built in AMP training](https://pytorch.org/docs/stable/notes/amp_examples.html).

## Feature Evaluation

Feature evaluation is done using a linear model protocol. 

First, we learned features using SimCLR on the ```STL10 unsupervised``` set. Then, we train a linear classifier on top of the frozen features from SimCLR. The linear model is trained on features extracted from the ```STL10 train``` set and evaluated on the ```STL10 test``` set. 

Check the [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://github.com/sthalles/SimCLR/blob/simclr-refactor/feature_eval/mini_batch_logistic_regression_evaluator.ipynb) notebook for reproducibility.

Note that SimCLR benefits from **longer training**.

| Linear Classification      | Dataset | Feature Extractor | Architecture                                                                    | Feature dimensionality | Projection Head dimensionality | Epochs | Top1 % |

|----------------------------|---------|-------------------|---------------------------------------------------------------------------------|------------------------|--------------------------------|--------|--------|

| Logistic Regression (Adam) | STL10   | SimCLR            | [ResNet-18](https://drive.google.com/open?id=14_nH2FkyKbt61cieQDiSbBVNP8-gtwgF) | 512                    | 128                            | 100    | 74.45  |

| Logistic Regression (Adam) | CIFAR10 | SimCLR            | [ResNet-18](https://drive.google.com/open?id=1lc2aoVtrAetGn0PnTkOyFzPCIucOJq7C) | 512                    | 128                            | 100    | 69.82  |

| Logistic Regression (Adam) | STL10   | SimCLR            | [ResNet-50](https://drive.google.com/open?id=1ByTKAUsdm_X7tLcii6oAEl5qFRqRMZSu) | 2048                   | 128                            | 50     | 70.075 |