scaplen-lab-drosophila-segmentation

Scaplen Lab Drosophila Segmentation

Example output on a real frame: raw footage, then stage 1 whole-fly segmentation, then stage 2 body parts (blue = head, green = thorax, red = abdomen).

Two YOLO26m instance-segmentation models for tracking Drosophila

Models

• stage1_wholefly_seg.pt: whole-fly segmentation, one class (fly). Finds and segments every fly in the full frame.
• stage2_bodyparts_seg.pt: body-part segmentation (head, thorax, abdomen), run on single-fly crops taken from stage 1.

Both are YOLO26m (medium). Validation mask mAP@50 is about 0.995 for each.

Usage

from ultralytics import YOLO
model = YOLO("stage1_wholefly_seg.pt")
results = model.predict("frame.png")

Full pipeline, training scripts, datasets, and analysis code are on GitHub: https://github.com/ciij05/scaplen-lab

Built by Christian Khoury, Scaplen Lab, 2026.