A deep learning approach to tracking the apex of a moving plant.
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.
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.
- Obtain data from Plant Tracer homepage.
- Clone this repository.
- This project uses
condaenvironment. Create the conda virtual environment usingconda env create -f environment.yaml. - Modify the
run.pyfile and run it to start the training procedure. This project uses Comet for all visualizations. Add your comet API key in therun.pyfile to see visualizations. - The models are saved in
modelsfolder and validation and testing results are stored in thelogsdirectory and can be visualized with theviz.pyfile.
This repository contains the code used for submission of this Report.
The output from validation and testing of the model can be seen below.
Plant-Tracer is released under Apache License 2.0.
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.