https://github.com/sjenni/LearningToSpotArtifacts

Self-Supervised Feature Learning by Learning to Spot Artifacts. In CVPR, 2018.
https://github.com/sjenni/LearningToSpotArtifacts

alexnet classification compter-vision deep-learning feature-learning machine-learning self-supervised-learning stl-10 tensorflow unsupervised-learning

Last synced: 21 days ago
JSON representation

Self-Supervised Feature Learning by Learning to Spot Artifacts. In CVPR, 2018.

• Host: GitHub
• URL: https://github.com/sjenni/LearningToSpotArtifacts
• Owner: sjenni
• License: mit
• Created: 2018-05-06T10:59:42.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2023-03-24T22:43:59.000Z (over 2 years ago)
• Last Synced: 2024-07-05T15:52:36.844Z (over 1 year ago)
• Topics: alexnet, classification, compter-vision, deep-learning, feature-learning, machine-learning, self-supervised-learning, stl-10, tensorflow, unsupervised-learning
• Language: Python
• Homepage: https://sjenni.github.io/LearningToSpotArtifacts/
• Size: 4.98 MB
• Stars: 18
• Watchers: 3
• Forks: 1
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-self-supervised-learning - [code
• awesome-self-supervised-learning - [code

README

          
# Self-Supervised Feature Learning by Learning to Spot Artifacts [[Project Page]](https://sjenni.github.io/LearningToSpotArtifacts/) 

This repository contains demo code of our CVPR2018 [paper](https://arxiv.org/abs/1806.05024). It contains code for unsupervised training on the unlabeled training set of STL-10 and code for supervised finetuning and evaluation on the labeled datasets. 

## Requirements

The code is based on Python 2.7 and tensorflow 1.12. See requirements.txt for all required packages.

## How to use it

### 1. Setup

- Set the paths to the data and log directories in **globals.py**.

- Run **init_datasets.py** to download and convert the STL-10 dataset.

### 2. Unsupervised Training

- To pre-train the autoencoder run **train_autoencoder_stl10.py**

- To train the classifier and the repair network run **train_stl10.py**

### 3. Transfer & Evaluation

- To finetune the learnt representations run  **fine_tune_stl10.py**

- To evaluate the finetuned classifier run **test_classifier_stl10.py**