Try to 'trick' the model, i.e. tell it one of the dataset has no object, but it actually does have data, so both datasets being fed in would be data for an object

ofithcheallaigh / masters_project

This repo is for my MSc in AI research project

0 stars 0 forks source link

Try to 'trick' the model, i.e. tell it one of the dataset has no object, but it actually does have data, so both datasets being fed in would be data for an object #26

Closed ofithcheallaigh closed 1 year ago

ofithcheallaigh commented 1 year ago

This is basically an idea I had to try and confirm that the 100% accuracy scores I was getting were correct. I wanted to find a way to make my models fall over.

To to this, I will feed in data from two object data sets, but tell the system that one of those data sets has no object. This should induce some poor performance.

ofithcheallaigh commented 1 year ago

I did this appraoch for a grid-wise analysis, and for a full data set binary analysis (is there an object or not).

The results showed that the results got worse for the grid-wise analysis, and worse again for the binary analysis:

ofithcheallaigh commented 1 year ago

With this, I think I am happy enough with the results, and have some more confidence the models are working correctly.