This code presents a Siamese Neural networks assessment for different embedded models using the QMUL-OpenLogo dataset, following the paper: One Shot Logo Recognition Based on Siamese Neural Networks.

• Python 3.x
• Numpy
• OpenCV
• Pytorch
• Matplotlib
• PIL
• QMUL-OpenLogo dataset (https://qmul-openlogo.github.io/)

Use the misc/data_prep.py script to preprocess the QMUL-OpenLogo dataset (crop and data split) by defining the python openlogo_path, python train_dir and python test_dir variables. Set the python params/config.py file to define the architecture to train and training parameters. Run the python main.py file to train/test the defined configuration.

https://dl.acm.org/doi/abs/10.1145/3372278.3390734

(Showing the resulting dissimilarity metric for each pair of images)

- d = 0.014

- d = 0.344

- d = 0.837

- d = 0.397

- d = 0.999