Context

This dataset was created for a final year university project. The aim was to classify and segment items that are on the tray. Images were annotated manually using Photoshop.

Content

The “TrayDataset” folder contains 4 additional folders:

• XTrain – 1241 total images, 17 unique trays, other images are rotated/flipped/wrapped versions of the unique trays.
• yTrain - truth labels for XTrain
• XTest – 8 unique images, the trays in the testing set is not present in the training set.
• yTest - truth labels for XTest

Scoring/Our results:

As part of this project 2 scoring matrices were used:

• IoU (jaccard_score) – Implementation link
• SSC – Custom matrices, designed to evaluate classification accuracy based on segmentation. Also see noise reduction that is preformed prior to calculating SSC. Implementation link – SSC, Noise reduction.

Our results Using Unet (implementation), SSC: 0.9464, IoU: .9226

Inspiration

A lot of code on the preparation/our project can be found here https://github.com/thezaza101/Meal-Compliance-Project

Errors

While i've done error checking it is possible that errors exist, please contact me if you find any and i will update the data set.