This is a MSc individual project to enhance the visual detection capabilities of USVs by utilizing Bayesian SegNet. This project aims to enhance the reliability of USVs in complex and dynamic marine environments despite the scarcity of marine datasets.
Bayesian SegNet on MaSTr1325 dataset
Bayesian SegNet on OASIs dataset
gantt
dateFormat YYYY-MM-DD
title PROJ00114PG Bayesian deep learning based semantic segmentation for marine environments
Mataining this Gantt Chart [Z. Ye] : crit, 2024-03-01, 2024-09-16
section Preliminary Preparation
Literature Review: activate, 2024-03-01, 2024-07-01
Outline Plan: done, crit, milestone, 2024-03-21, 0d
Project Planning: done, 2024-03-21, 2024-04-30
Project Initialization: done, crit, milestone, 2024-05-18, 0d
section Design & Development
BNN on Simple Regression Problem: done, 2024-05-18, 2024-05-28
WR - 20240528: done, milestone, 2024-05-28, 0d
Relationship between BNN and Bayesian Inference: done, 2024-05-28, 2024-06-04
WR - 20240604: done, milestone, 2024-06-04, 0d
Bayesian CNN on simple image segmentation: activate, 2024-06-04, 14d
Term 3 Progress Report: done, crit, milestone, 2024-06-10, 0d
WR - 20240611: milestone, 2024-06-11, 0d
WR - 20240618: milestone, 2024-06-18, 0d
Bayesian CNN on MaSTr1325: 2024-06-18, 21d
WR - 20240625: milestone, 2024-06-25, 0d
WR - 20240702: milestone, 2024-07-02, 0d
Term 4 Progress report: crit, milestone, 2024-07-08, 0d
WR - 20240709: milestone, 2024-07-09, 0d
section Testing
Overall Testing and Parameter Adjustment: 2024-07-09, 2024-08-20
Testing on Prediction Uncertainty: TPU, 2024-07-09, 15d
WR - 20240716: milestone, 2024-07-16, 0d
WR - 20240723: milestone, 2024-07-23, 0d
Testing on Dataset Noise: TDN, after TPU, 15d
WR - 20240730: milestone, 2024-07-30, 0d
WR - 20240806: milestone, 2024-08-06, 0d
Testing on Proir: TP, after TDN, 15d
WR - 20240813: milestone, 2024-08-13, 0d
WR - 20240820: milestone, 2024-08-20, 0d
Testing Complement: crit, milestone, 2024-08-20, 0d
section Thesis Writing
Thesis Writing: 2024-08-09, 2024-09-02
WR - 20240827: milestone, 2024-08-27, 0d
Thesis Submission: crit, milestone, 2024-09-02, 0d
Oral Examination Preparation : 2024-09-02, 2024-09-16
WR - 20240903: milestone, 2024-09-03, 0d
Oral Presentation Material Submission: crit, milestone, 2024-09-06, 0d
Oral Examination: crit, milestone, 2024-09-16, 0d
The instructions of downloading datasets are illistruted in the link.
The corresponding *.tar.gz files of datasets are provided as well in Dataset file.
Conda environment
conda create -n <your-env-name> python=3.10
conda activate <your-env-name>
pip install -r requirements.txt
python main.py --epoch 200 --arch segnet --batch_size 4 --dataset MaSTr1325 --action train&test
python train.py --data-path MaSTr1325 --epoch 1000
Performance on MaSTr1325 dataset
| Architecture | Pr (%) | Re (%) | F1 (%) |
|---|---|---|---|
| Bayesian SegNet | 81.2 | 97.8 | 87.8 |
| SegNet | 79.9 | 87.5 | 81.3 |
| PSPNet | 82.1 | 50.8 | 62.8 |
| U-Net | 10.2 | 88.6 | 18.3 |
Performance on OASIs dataset
| Architecture | Evaluation Dataset | Type | Pr (%) | Re (%) | F1 (%) |
|---|---|---|---|---|---|
| Bayesian SegNet | OASIs | 1 | 68.73 | 84.31 | 72.67 |
| 2 | 47.27 | 95.74 | 64.36 | ||
| 3 | 65.20 | 88.89 | 73.52 | ||
| SegNet | OASIs | 1 | 28.27 | 65.89 | 33.93 |
| 2 | 1.69 | 97.17 | 3.32 | ||
| 3 | 10.09 | 98.97 | 17.82 | ||
| SegNet | SMD | -- | 31.2 | 76.3 | 44.3 |