Confidence differences in MLfK vs. Scratch

Aubrey Sherman

unread,

Oct 16, 2025, 10:52:09 AMOct 16

to Machine Learning for Kids

For image recognition models:

I noticed a difference in labeling and confidence when I test an image within the MLfK platform and then on Scratch. In the screenshots, you'll see the label and confidence for the same image -- rock is the accurate label. Why might this discrepancy be happening, and how can I get these results to be consistent with each other?

Thanks!

Dale Lane

unread,

Oct 16, 2025, 10:53:26 AMOct 16

to Machine Learning for Kids

This is almost certainly because you're not actually testing the same image in the two places.

Can you describe how you are doing this please?

D

Aubrey Sherman

unread,

Oct 16, 2025, 11:17:38 AMOct 16

to Machine Learning for Kids

On each platform, I took a photo with the webcam with the same background and tried to match the same pose/positioning/lighting as closely as possible. I did these in immediate succession.

Dale Lane

unread,

Oct 16, 2025, 11:18:13 AMOct 16

to Machine Learning for Kids

Would you be willing to share your Scratch code, please?

Reply all

Reply to author

Forward