search

bgorman3 / ITSC-3990-AI-Art-Authentication

New Research Project for ITSC 3990 Using AI to Authenticate Art
1 stars 0 forks source link

Fourth Task #11

Closed btdobbs closed 1 week ago

btdobbs commented 2 weeks ago

Evaluate the model using 136 AI-generated images to match the size of the Monet test set.

btdobbs commented 2 weeks ago

It may be helpful to generate all AI images from the same source at first. Multiple sources is an interesting test too.

bgorman3 commented 2 weeks ago

https://docs.google.com/document/d/1W-03tnIgmOoi79_zDj9IQtf_kzjcjmkKt4A23kNet1I/edit?usp=sharing I've completed tasks 1-4 but some notes I don't have the 136 ai pictures from one source I think ill ask you for help getting that and when I evaluated the AI dataset I didn't use any monet so I might change it to do that in the future but my model is

Combined Test Set Performance: Accuracy: 0.9632 Precision: 0.9437 Recall: 0.9853 F1 Score: 0.9640 accurate and I don't have any graphs like comparing everything or confusion matrices yet for my poster I will get those tomorrow

btdobbs commented 2 weeks ago

Nice work! I think these results look good and you have a minimum of what is needed for your poster! I was expecting the binary classification to be higher than the multi-classification work. While the validation accuracy is a little higher than the test accuracy, I think it is a reasonable gap. Let's discuss more on Tuesday, but please let me know if you have further questions before then.

bgorman3 commented 2 weeks ago

Which multiclassification work are you referring to? Are you talking about the results from your experiment?

btdobbs commented 2 weeks ago

yes