Vent: my little image sorter choked on 8,000 photos from my cousin's wedding in Buffalo
I set up a small open source model to sort my cousin's wedding photos from Buffalo last weekend, about 8,000 shots, and it labeled half the reception as cake cutting. I was ready to call AI sorting a party trick until I cropped the faces tighter and bumped the batch size down, and it got way more accurate. Still can't figure out if that's the model being picky or me feeding it junk. Anyone got a trick for tuning batch size on this stuff, or is it just trial and error?
Sounds like you just found your answer by messing with the settings, so not sure there's a big mystery here. Cropping faces and dropping batch size is just basic tuning, not some magic trick. Run a few batches, see what works, and move on lol.