i am using pretrained EfficientNet-B0 as the image encoder and 5 fold cross validation
OOF RMSE: 1.50928179892321
LB 0.978123577
Also any ideas for potential improvment
My best efficientnet b0 is 0.98270215 LB 1.4221 4-fold CV
Forget the other chat, but agree with a comment re the number of samples in test (it's not much!). So there will be a big difference here and probably a big shake up in the end. Seems like a gamble to me haha
EfficientNet-B0 is lightweight. Good baseline, but not optimal.
Trees Hate You Try:
Even a single upgrade here can outperform heavy tuning on B0.
'/%3e%3cpath fill='%23fff' d='m8.751-17.959 10.11 11.373L3.997-9.844l13.94-6.1-7.692 13.129z'/%3e%3c/svg%3e){kind=small}
My best efficientnet b0 is 0.98270215 LB 1.4221 4-fold CV
Forget the other chat, but agree with a comment re the number of samples in test (it's not much!). So there will be a big difference here and probably a big shake up in the end. Seems like a gamble to me haha
EfficientNet-B0 is lightweight. Good baseline, but not optimal.
Trees Hate You Try:
Even a single upgrade here can outperform heavy tuning on B0.