Single Particle Tracking using Deep learning

Python, Deep Learning, Transformer, Computer Vision

Pertaining to our previous project of analysing single-particle tracking data, there is also a lack of a universally accepted algorithm for tracking particles. All the software available for this task either require knowledge of some specific programming language or have highly lethal user interfaces. Alongside that, these algorithms heavily rely on the user for tuning the parameters each time they need to track their data. This creates a user bias and makes it hard for others to reproduce the same results, which is of utter significance in biology.

Recognizing these problems, we planned to develop a particle tracker (multi-object) based on neural network architectures. Neural networks were chosen owing to the fact that once trained, they can be used on any machine and also require minimal to no input from the end-user, which makes the reproducibility of results much more straightforward.

This project started out of sheer interest during my bachelor's degree and now adopted as my master's thesis project. We are trying to use LSTMs and CNNs in conjunction to get better results. We also plan to use transformer neural networks (new hot thing) at some point to achieve SOTA results in this task. Once developed, this model can be used by anyone across the globe directly or can be trained on custom datasets.

• Date: January 2020-Present