Fruits360

Simple NN for fruits classification based on Fruits360 dataset

Description

Implementation of simple nn with convolutional layers and last linear layer with softmax to predict class of the image.

Some notes

1. Flattener layer used to change input from conv layers to Fully-Connected\Linear.
2. There are no biases in convolutional layer, because of batch normalization layers. BN has biases inside.
3. Trained on GPU in Google Colab.
4. There is no need to use grid search, because it is easy to achieve training convergence without much experiments. Just make higher L2 to obtain better result on test.