Pytorch-SiamFC

Why we included this project

SiamFC is the classic fully-convolutional Siamese tracker from the 2016 paper, and this repo gives you a readable PyTorch port of it. Instead of shipping only the inference path, it covers the whole pipeline: you can train the embedding network on ImageNet VID to learn a patch similarity metric, run the tracker on new videos, and inspect the correlation maps that show where the target is. The code is organized into training, tracking, and visualization parts, with experiment folders, parameter files, and TensorBoard logging that make it easy to tweak settings and compare runs. The included baseline pretrained weights help you get results without training from scratch, which makes it a good starting point for researchers and students who want to see how Siamese trackers actually work. One caveat: the project has not been updated since 2020.