3
My AI model finally learned to tell cats from dogs after 4 failed attempts
I was sitting in my home office in Portland last Tuesday, watching my training data go through the same mislabeling loop for the third time. Every single photo of a corgi was coming back as a tabby cat, and I couldn't figure out why. Turns out I had accidentally swapped the training folders when I set up the script on Monday night. After fixing the path and rerunning it, the accuracy jumped from 45% to 92% in one go. Has anyone else had a dumb folder mixup blow up a whole test run?
2 comments
Log in to join the discussion
Log In2 Comments
jennifer_west29d agoTop Commenter
Oh man, I'd probably swap my lunch with the dog food next time!
6
And that 45% should have been a HUGE red flag right there. I had a similar folder swap with my plant dataset and it kept labeling succulents as "rocks" for two days. Nothing like a good folder structure to remind you how much you don't know.
1