PCA-ResFeats

We propose methods to improve the efficiency of image analysis by using local image descriptors in the form of appropriately formed semantic features extracted from the ResNet-50 deep neural network – PCA-ResFeats.

Instalation

First, download the repository:

$ git clone git@github.com:Stanislaw-Lazewski/PCA-ResFeats.git
$ cd PCA-ResFeats

Second, create conda envirement:

$ conda env create -n pca-resfeats --file environment.yml
$ conda activate pca-resfeats

Demo

Download Shells and Pebbles dataset available at: https://www.kaggle.com/datasets/vencerlanz09/shells-or-pebbles-an-image-classification-dataset, unzip and then place it in the data_demo/Shells_and_Pebbles directory.

Run src/demo.py:

python3 src/demo.py

The files .npy in the pca-resfeats folder contain the extracted PCA-ResFeats. File pca-resfeats/classification_report.txt contains the classification results.

Experiments

Data preparation

Prepare the datasets from which you want to extract PCA-ResFeats and place them in the data directory. Each class should be placed in its own subdirectory. The datasets I have used are:

data
├── Caltech101
├── Caltech256
├── Catordog
├── EuroSAT
├── Flowers
└── MLC2008

Running experiments

Run src/main.py to repeat my experiments or adapt the function calls to your own datasets:

python3 src/main.py

Experiments results

The results of the experiments are available here: https://aghedupl-my.sharepoint.com/:f:/g/personal/slazewsk_agh_edu_pl/EpE8LQmXBIpDpVU44eYjuMYBMBCdbs3gO0EtEz9LIfo02w?e=TaumAv

License

MIT