1) with custom loss and redirects augmentation
2) with custom loss and without redirects augmentation
3) without custom loss and with redirects augmentation (acts as baseline performance)
Measure metrics and training curves to see stability of training, train with higher epochs to see learning and improvement in recall
MihailSalnikov
commented
1 year ago
Run training (same hyper parameters):
1) with custom loss and redirects augmentation 2) with custom loss and without redirects augmentation 3) without custom loss and with redirects augmentation (acts as baseline performance)
Measure metrics and training curves to see stability of training, train with higher epochs to see learning and improvement in recall