AdamHines/samurai-beetracker

Description

This repository contains the fine tuned SAM2.1 model used to segment and track bees using the SAMURAI zero-shot motion tracking system. The model was fine tuned on 100 images from experimental recordings of bees with manually annotated masks. The images and associated ground truth masks are available in sam2_ft.zip. We also provide the code used to finetune the model.

The full tracking code can be found at https://github.com/AdamDHines/samurai-beetracking.

Files

• sam2.1_hiera_large.pt -- The fine tuned SAM2.1 model.
• sam2_ft.zip -- Zip file containing raw images, segmentation masks, and bounding box annotations for fine tuning.
• finetune-sam2.py Python script containing the fine tuning code using the segmentation masks.