aakaashjois/Plant-Tracer

A deep learning approach to tracking the apex of a moving plant

Plant Tracer

A deep learning approach to tracking the apex of a moving plant.

Introduction

Plant Tracer is an app designed to enable analysis of plant movement from time-lapse videos. This repository contains a small part of the complete project. Here, I try to track the movement of the plant using a deep learning model to make it robust and work on videos with occlusion while tracking plants.

Architecture

The architecture follows the architecture in GOTURN. The CaffeNet pretrained on CIFAR-100 has been replaced with AlexNet pretrained on ImageNet. The architecture of the model is shown below.

Instructions

1. Obtain data from Plant Tracer homepage.
2. Clone this repository.
3. This project uses conda environment. Create the conda virtual environment using conda env create -f environment.yaml.
4. Modify the run.py file and run it to start the training procedure. This project uses Comet for all visualizations. Add your comet API key in the run.py file to see visualizations.
5. The models are saved in models folder and validation and testing results are stored in the logs directory and can be visualized with the viz.py file.

Report

This repository contains the code used for submission of this Report.

Result

The output from validation and testing of the model can be seen below.

Validation

Tracking Test

License

Plant-Tracer is released under Apache License 2.0.

Author

This repository and the approach has been created by Aakaash Jois.

The complete Plant Tracer application has multiple authors and details related to that can be found on Plant Tracer homepage.