Closed hankyul2 closed 2 years ago
I am training EfficientNetV2 without pretrained weights on CIFAR-100 and... thank you for this! Your optimizer and scheduler settings are the only ones that can make the loss converge!
Because i am doing from scratch and not finetuning, my learning rate is 0.1 and epochs=100. It would be good if you can share the finetuning results.
Hi @iamanigeeit
Thank you for your interesting question. I already shared the finetuning results such as checkpoint and training logs on the experiment result section. If you click the tf.dev link, you can find the full val top-1/5 accuracy using tensorboard.dev shown in the below graph.
Thank you.
Hankyul.
Run EfficientNet on cifar10/100
Model: EfficientNet-S, M, L, XL
Dataset: Cifar10
[x] Prepare cifar setup
[x] Prepare ImageNet Pretraining setup
[x] Run Model on Cifar10
[x] Run Model on Cifar100