Loading…
29
TIL my edge detection model kept calling my neighbors cat a raccoon, took 11 days to fix
Back in March I trained a small image model on 800 backyard photos to tell raccoons from cats so my trash cam would stop pinging me at 2am. The thing hit 94% on my test set but every single night it flagged the cat, and it turned out my training shots were all taken at dusk so the model just learned 'dark and blurry means raccoon'. Anyone else get burned by their own dataset before they ever blamed the model?
0 comments
Log in to join the discussion
Log In0 Comments
No comments yet
Be the first to share your thoughts on this discussion.