CSE488 - Group 6 - Cell Segmentation and Evaluation Project

Project Overview

Dataset(s):
- Fluo-N2DH-GOWT1
Team roles:
- Segmentation Lead: Jeremy Lunka
- Evaluation Lead: Jeremy Lunka
Summary:
- Group 6: Jeremy Lunka
- This cell-segmentation project implements a segmentation and evaluation pipeline on the Fluo-N2DH-GOWT1 dataset, using classical models and the official Cell Tracking Challenge evaluation tools.

Segmentation Results

Method	Dataset	Track	Mean IoU	Notes
Classical SVM	Fluo-N2DH-GOWT1	01	0.73	RBF kernel, window = 5
CNN Baseline	Fluo-N2DH-GOWT1	02	0.81	Lightweight U-Net

Describe preprocessing, feature engineering, and model choices.
Link to notebooks or scripts that reproduce each row.

Demo Videos / GIFs

MP4/GIF assets should be committed under docs/assets/. The example below assumes a file at docs/assets/segmentation_demo.mp4.

Reproduction Checklist

conda env create -f environment.yml
conda activate cse488-cell-tracking
python scripts/setup_data.py Fluo-N2DH-GOWT1
python scripts/train_classical.py Fluo-N2DH-GOWT1 --model svm
python scripts/eval_classical.py Fluo-N2DH-GOWT1 --model svm
Additional steps for other classical models and experiments.

Lessons Learned

Summarize key findings, segmentation challenges, and what you would try next (e.g., improved features, different classical models, or deep architectures such as U-Net).