My model doesn't predict well (I guess)

I pretty much just copypasted the cells from the 00-is-it-a-bird-creating-a-model-from-your-own-data notebook and substituted with my own values. That was the result for my model:

Screenshot 2023-05-18 at 9.36.49 am

Quite a difference from the original notebook:

Screenshot 2023-05-18 at 9.37.50 am

Is this because my model isn’t predicting something as clear as the difference between a bird and a forest? My model predicts the difference between a kangaroo and a wallaby.

And is this really all the work we should do in the first class? Just run the cell ourselves and try our own images so that we just get used to it?

Yes. Distinguishing a kangaroo from a wallaby is (presumably) more difficult for a DL model than distinguishing a bird from a forest, which explains your model’s decreased accuracy. (My first model performed even worse. :sweat_smile:)

Down the line you’ll learn techniques on how to improve the accuracy of your model. But yeah, I think the purpose of Lesson 1 is to introduce one to the world of Deep Leaning and “show them the whole game,” gradually going deeper with subsequent lessons.

1 Like