An Experiment on Animal Re-Identification from Video

1. Data

The data is from repository: Animal Re-Identification from Video.

2. Code

The code is divided in two parts, the first is in MATLAB with the feature extractions, the second is in Python with more feature extraction, the deeplearning and lazypredict.

2.1 MATLAB

For complete explanation, see matlab folder. All features can be extracted using Extract_AE_Features.m and PrepareData.m scripts.

• AE (autoencoder) features: Extract_AE_Features.m
• HOG features: extractHOGFeatures
• LBP features: extractLBPFeatures
• RGB features: get_rgb_features.m

2.2 Python

• MN2 (ImageNetV2) features: In the file functions.py it is the extractMobilNetfeatures function. Recives the output of the select_half_video and return the dataset with MobileNetV2 features.
• Deeplearning: In the Jupyter Notebook DeepLearningClassification the Convolutional Neural Network (CNN) and the Transfer Learning with MobileNetV2 can be executed. The variable training_fold choose which half of the video will be used as training.
• lazypredict: In the Jupyter Notebook lazypredict the experiment with different classifiers can be executed.

Acknowledgement

• This work is supported by the UKRI Centre for Doctoral Training in Artificial Intelligence, Machine Learning and Advanced Computing (AIMLAC), funded by grant EP/S023992/1.
• This work is also supported by the Junta de Castilla León under project BU055P20 (JCyL/FEDER, UE), the Ministry of Science and Innovation under project PID2020-119894GB-I00 co-financed through European Union FEDER funds, and the Ministry of Universities under mobility grant PRX21/00638.
• J.L. Garrido-Labrador is supported through Consejería de Educación of the Junta de Castilla y León and the European Social Fund through a pre-doctoral grant (EDU/875/2021).
• I. Ramos-Perez is supported by the predoctoral grant (BDNS 510149) awarded by the Universidad de Burgos, Spain.