Repositories
Unsupervised Distance Learning Framework - UDLF
Repository: https://github.com/UDLF/UDLF
A framework of unsupervised distance learning methods for image and multimedia
retrieval tasks. Currently, ten different unsupervised learning methods are implemented :
BFS-Tree
Breadth-First Search Tree
https://github.com/UDLF/UDLF/blob/master/src/Methods/BFSTree.hpp
https://doi.org/10.1016/j.patcog.2020.107666
RDPAC
Rank-Based Diffusion Process with Assured Convergence
https://github.com/UDLF/UDLF/blob/master/src/Methods/RDPAC.cpp
https://doi.org/10.3390/jimaging7030049
LHRR
Log-based Hypergraph of Ranking References
https://github.com/UDLF/UDLF/blob/master/src/Methods/LHRR.cpp
https://doi.org/10.1109/TIP.2019.2920526
ContextRR
Context Re-Ranking
https://github.com/UDLF/UDLF/blob/master/src/Methods/Contextrr.cpp
https://dl.acm.org/doi/10.5555/1948207.1948291
CG
Correlation Graph
https://github.com/UDLF/UDLF/blob/master/src/Methods/CorrelationGraph.cpp
https://doi.org/10.1016/j.neucom.2016.03.081
CPRR
Cartesian Product of Ranking References
https://github.com/UDLF/UDLF/blob/master/src/Methods/Cprr.cpp
https://doi.org/10.1109/SIBGRAPI.2016.042
Rk Graph Dist
Ranked List Graph Distance
https://github.com/UDLF/UDLF/blob/master/src/Methods/RkGraph.cpp
https://doi.org/10.1016/j.patrec.2016.05.021
ReckNNGraph
Reciprocal kNN (k-Nearest-Neighbor) Graph
https://github.com/UDLF/UDLF/blob/master/src/Methods/ReckNNGraph.cpp
https://doi.org/10.1016/j.imavis.2013.12.009
RL-Recom
Ranked Lists Recommendation
https://github.com/UDLF/UDLF/blob/master/src/Methods/RlRecom.cpp​
https://doi.org/10.1145/2671188.2749336
RL-Sim
Ranked Lists Similarities
https://github.com/UDLF/UDLF/blob/master/src/Methods/RlSim.cpp
https://doi.org/10.1145/2671188.2749335
​
Feature Augmentation based on Manifold Ranking and LSTM for Image Classification
Published paper: https://doi.org/10.1016/j.eswa.2022.118995
Repository: https://github.com/vanessaferrero/LSTM
Code Ocean: https://doi.org/10.24433/CO.3814711.v1
​
This code is based on an LSTM - Long Short Term Memory - network implementation using Python and Keras. It was also based on the related repositories:
LSTM for MNIST
https://github.com/ar-ms/lstm-mnist
UDLF framework
Oxford Flowers 17 dataset
https://www.robots.ox.ac.uk/~vgg/data/flowers/17/index.html
LHRR manifold learning paper
https://ieeexplore.ieee.org/document/8733193
​
Weakly Supervised Experiments Framework (WSEF)
Repository: https://github.com/UDLF/WSEF
Published paper: https://doi.org/10.1109/ICPR48806.2021.9412596
It proposes a rank-based model to exploit contextual information encoded in the unlabeled data
in order to perform label expansion and execute a weakly supervised classification.
​
Related Datasets
The links for the most commonly utilized test datasets used in the source code
related publications.
CIFAR10
https://www.cs.toronto.edu/~kriz/cifar.html
A public image dataset of 60,000 images, with 50,000 in the training set and 10,000 in the test set.
The dataset is composed of natural images from 10 different classes.
Linnaeus5
http://chaladze.com/l5/
A dataset composed of 8,000 images divided into 5 classes (berry, bird, dog, flower, and other), with 6,000 training images (1,200 training images per class) and 2,000 test images (400 test images per class).
Oxford Flowers 17
https://www.robots.ox.ac.uk/~vgg/data/flowers/17/
A well-known dataset for common UK flower classification evaluation is composed of 17 categories of flowers divided into 80 images, totaling 1,360 images.
Oxford Flowers 102
https://www.robots.ox.ac.uk/~vgg/data/flowers/102/
A larger dataset for flower classification consists of 102 flower categories. Each class consists of between 40 and 258 images, totaling 8,189 images.
Stanford Dogs
http://vision.stanford.edu/aditya86/ImageNetDogs/
The dataset contains images of 120 breeds of dogs from around the world. This dataset has been built using images and annotation from ImageNet for the task of fine-grained image categorization. It includes 12,000 training images and 8,580 test images, totaling 20,580 images.
MPEG-7
https://dabi.temple.edu/external/shape/MPEG7/dataset.html
There are 1,400 shape images divided into 70 classes. This dataset is commonly used on evaluation of unsupervised post-processing methods for shape retrieval.
Corel5k
https://rdrr.io/cran/mldr.datasets/man/corel5k.html
The dataset includes diverse scene content such as fireworks, bark, microscopy images, tiles, trees, and others. It is composed of 50 categories with 100 images for each class.
MNIST
http://yann.lecun.com/exdb/mnist/
Handwritten digit database composed by 60,000 training and 10,000 test images of ten different classes of numbers from 0 to 9.
Brodatz
https://www.ux.uis.no/~tranden/brodatz.html
The 112 texture images given in the Brodatz album have different background intensities.
