YOLOv3-for-Particle-Tracking

This repository contains the code for the Bachelor's thesis: Karakterisering och spårning av nanopartiklar med djupinlärning (eng. Characterization and tracking of nanoparticles using deep learning) by Arash Darakhsh, Edvin Johansson, Simon Nilsson, Sanna Persson and Rickard Ström at Chalmers University of Technology

Link to thesis: https://odr.chalmers.se/handle/20.500.12380/304358

How to use the code

Download the repository and weights

To clone the repository run:

git clone https://github.com/Deep-learning-for-particle-tracking/YOLOv3-for-Particle-Tracking.git

The weights are available at this link at Kaggle. The source code and model weights can also be downloaded in the release at Github. In the data-folder in the release there are two example experimental images.

Install requirements

For the installation of PyTorch it is best to check out the PyTorch website especially if you want a specific CUDA version.

pip install requirements.txt

Inference with the model

Place the weights in the model folder and the image for inference in npy-format in the data-folder. Run

python detect_on_patches.py

For information on flags and arguments run:

python detect_on_patches.py --help

There are a couple of example experimental images in the folder data which you can test the model on.

Train the model

Change the configuration for training in the config.py file or run

python train.py --help

to read about the training parameters to input them in the terminal. To train the model run

python train.py 

A few examples of how to structure the training data is also found in the training_data folder.

Simulate images

The code for simulating the images is found in the simulation folder.