Mastering Computer Vision with TensorFlow 2.x

Siamese network-based tracking

Siamese network-based object tracking was proposed in 2016 by Luca Bertinetto, Jack Valmadre, Joao F. Henriques, Andrea Vedaldi, and Philip H. S. Torr in their landmark paper Fully-Convolutional Siamese Networks for Object Tracking. The details of the paper can be found at https://arxiv.org/abs/1606.09549.

In this paper, the authors trained a deep convolution network to develop a similarity function offline and then applied this to real-time object tracking. The similarity function is a Siamese CNN that compares a test bounding box to a training bounding box (ground truth) and returns a high score. If the two bounding boxes contain the same object and a low score, then the objects are different.

A Siamese network ...