ghostk

5 ms·

1. 1.▲
   by ghostk 25d ago

   type shit Louis the XVI would ponder about
2. 2.▲
   by ghostk 4y ago

   Without changing the architecture in a significant manner/using more training data, there does come a point where adding more parameters will result in no gain (and sometimes even worse results). There's also a consequence of perf
3. 3.▲
   by ghostk 4y ago

   Probably not much at all. They used a pretty straightforward and easy to run model, EfficientNetV2. Something that doesn't give out the best results, but will offer very quick inference and train time.