Skip to content

CLEAR: Cumulative LEARning for one-shot one-class image recognition (CVPR 2018)

Notifications You must be signed in to change notification settings

JKozerawski/CLEAR-osoc

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

39 Commits
 
 
 
 
 
 
 
 

Repository files navigation

CLEAR: Cumulative LEARnining One Shot One Class Image Recognition

This repository contains the code (Caffe) for "CLEAR: Cumulative LEARnining One Shot One Class Image Recognition" paper (CVPR 2018) by Jedrzej Kozerawski and Matthew Turk.

Our CVPR poster is available here

Citation

@inproceedings{kozerawski2018clear,
  title={CLEAR: Cumulative LEARning for One-Shot One-Class Image Recognition},
  author={Kozerawski, Jedrzej and Turk, Matthew},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={3446--3455},
  year={2018}
}

Introduction

This work addresses the novel problem of one-shot one-class classification. The goal is to estimate a classification decision boundary for a novel class based on a single image example. Our method exploits transfer learning to model the transformation from a representation of the input, extracted by a Convolutional Neural Network, to a classification decision boundary. We use a deep neural network to learn this transformation from a large labelled dataset of images and their associated class decision boundaries generated from ImageNet, and then apply the learned decision boundary to classify subsequent query images. We tested our approach on several benchmark datasets and significantly outperformed the baseline methods.

Training pipeline

Testing pipeline

Usage

Dependencies

  • Caffe
  • Numpy
  • ILSVRC2012 (for training)

Train

  1. Extract features from all training images using the Inception V1 network (or other if preferred)
  2. Train One-Vs-Rest classifier for ever category using the extracted features (e.g. linear SVM)
  3. Select randomly training pairs:
    • input: feature extracted from an image
    • target output: parameters of One-Vs-Rest classifier (for category that the input image belongs to)
  4. Create HDF5 files to train the network.
  5. Using our train_val files (link below) train the network.

TO BE COMPLETED

Pretrained model

Tar package with Caffe model files is available here

Test

In the test.py file we use feature extraction script (using Inception V1 network) to do preprocessing of images. If you want to use our test.py file - please download the Inception V1 .caffemodel and deploy files available here. Make sure to point to the directory where the model is with the --inception_model_dir flag when using the test.py

To test the network just run our example:

python test.py --model_dir pathToTheCLEARNetwork --inception_model_dir pathToTheInceptionNetwork --image imagePath --imageR pathToTheRightImage

About

CLEAR: Cumulative LEARning for one-shot one-class image recognition (CVPR 2018)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages