dipuk0506 / SpinalNet

SpinalNet: Deep Neural Network with Gradual Input
171 stars 42 forks source link

EMNIST balanced ACC #6

Closed Bellman281 closed 2 years ago

Bellman281 commented 2 years ago

Dear Scholars.

I have trained the SpinalNet model on EMNIST BALANCED data. Unfortunately I could achieved 85.82 ACC after 100 epoch training. Would you please comment if there is a way to get better accuracy on this dataset? I have prepared the notebook as following for convenience:

https://github.com/Bellman281/Character_Recognition_Models

Any comments is highly appreciated.

dipuk0506 commented 2 years ago

Hi Bellman, Please use one script from the MNIST VGG folder. EMNIST digits or EMNIST letters. They are the high-performing ones. https://github.com/dipuk0506/SpinalNet/tree/master/MNIST_VGG

Also, you can practice transfer learning and get good performance in fewer epochs. For MNIST transfer learning, see the notebook: https://github.com/dipuk0506/SpinalNet/blob/master/Jupyter%20Notebooks/transfer-learning-on-fashion-mnist-dataset.ipynb