EnmmmmOvO/SeaTurtleID2022_9517Project

Model Card for Model ID

This repository contains FCN and FPN models trained using the split_open strategy from SeaTurtleID2022. The models have been uploaded to facilitate evaluation and review by the supervisor.

Model Details

Model Description

• Developed by: Jinghan Wang
• Language(s) (NLP): English
• License: MIT

Model Sources [optional]

• FPN: Segmentation Models Pytorch
• FCN: Pytorch

Uses

These models have been uploaded to facilitate review and evaluation for the COMP9517 Computer Vision project assignment at the University of New South Wales, T3 2024 semester.

Training Details

Training Data

SeaTurtleID2022 split_open: 'train'

Valid Data

SeaTurtleID2022 split_open: 'valid'

Testing Data, Factors & Metrics

Testing Data

SeaTurtleID2022 split_open: 'test'

Evaluation

FCN (Resnet101)
  Mean IoU: 0.9039
  Mean Accuracy: 0.9458

  Mean IoU of backdrop: 0.9932
  Mean Accuracy of backdrop: 0.9963

  Mean IoU of turtle: 0.9225
  Mean Accuracy of turtle: 0.9713

  Mean IoU of flipper: 0.8351
  Mean Accuracy of flipper: 0.8940

  Mean IoU of head: 0.8649
  Mean Accuracy of head: 0.9215

FPN (Resnet152):
  Mean IoU: 0.9042
  Mean Accuracy: 0.9440

  Mean IoU of backdrop: 0.9933
  Mean Accuracy of backdrop: 0.9966

  Mean IoU of turtle: 0.9242
  Mean Accuracy of turtle: 0.9708

  Mean IoU of flipper: 0.8317
  Mean Accuracy of flipper: 0.8899

  Mean IoU of head: 0.8677
  Mean Accuracy of head: 0.9187

Environmental Impact

• Cloud Provider: Colab