Comparisong of SIFT and CNN

This project compares using SIFT with color features and both trained and pretrained CNNs in a butterfly classification task.

The data_pipeline folder contains the different datasets used, implemented as PyTorch datasets and dataloaders. The models folder contains the baseline CNN used in the project. The training has all the code required to train the baseline CNN and the fine-tuned ImageNet classifier. The classifier has the scripts used to obtain results from trained CNNs and SIFT features with an SVM.

The butterfly dataset

This Butterfly-200 dataset used in this project is not included in the repo. You can download it from here, and move it to a folder called data in the root.