Config for best accuracy stage 1 model

ShadeAlsha / LTR-weight-balancing

CVPR 2022 - official implementation for "Long-Tailed Recognition via Weight Balancing" https://arxiv.org/abs/2203.14197

MIT License

118 stars 10 forks source link

Config for best accuracy stage 1 model #4

Closed CVHvn closed 2 years ago

CVHvn commented 2 years ago

Can you give me the configuration of the best model in stage 1. You write (and in demo 1) stage 1 with 46.7% accuracy but you load weights with 47.9% accuracy .

CVHvn commented 2 years ago

I see that you using test time augmentation (+ flip(3) / 2), sorry because I don't read carefully :(

ShadeAlsha commented 2 years ago

Thanks for you interest! I am glad that you figured it out. Also, note that we choose the model that had the best validation accuracy during training, not necessarily the model at the last epoch.