Weekly Updates 2025

Week 1 - Setup

Familiarized ourselves with the project and the progress that had been made in the past 5 years. Moreover, we figured out the next steps for the project, which were to run a control experiment on an empty ant habitat and to look for more confounding variables.

Week 2 - Linux Command Line and Bugs

We started learning the Linux Command Line and the existing data pipeline in order to get familiar with the working environment. We ran the code on the control data and discovered issues with the existing code, such as memory limitations and labelling problems.

Week 3 - Image Compression

We decided that the optimal course of action would be to rework the image compression algorithm that was present in the pipeline and switch from using numpy arrays and npz files to using PNG compression. This caused many problems due to the way the pipeline was structured, and the tar files were completely empty.

Week 4 - Debugging Pt 1

We progressed through the data pipeline step by step and debugged every step to work with the new PNG compression. We managed to get tar file creation working correctly with the right format, and are working on structuring the flatbin files to work correctly with the machine learning.

Week 5 - Debugging Pt 2

Finished png conversion code and continued debugging data pipeline and bin file conversion issues.

Week 6 - Debugging Pt 3

Finally found the error in the bin file conversion. Updated submodules and started running fixed code on control data.

Week 7 - Sampling Speed Up

The old pipeline was ~0.5 sec per sample. Using PNG compressions: ~0.07 per sample. Changing sampling structure: ~0.03 per sample. It is now much faster to process the millions of frames we have. We also fixed the bee_analysis segmentation labeling tool compatibility to work with both macOS and linux. We have also working on the paper.

Week 8 - Data Cropping + Saliency Maps

We worked on cropping the data to remove the empty space around the ants, which will help with the machine learning. We also updated the annotation tool to create visualizations using saliency maps.