fashionmnist

Solution of Kaggle's Fashion MNIST Image classification

Layer	Description	Shape
ConvA	32, 5x5, 1x1	(Batch_size, 28, 28, 1)
BatchNormA		(Batch_size, 28, 28, 32)
MaxPoolA	2x2, 2x2	(Batch_size, 28, 28, 32)
ConvB	64, 5x5, 1x1	(Batch_size, 14, 14, 32)
BatchNormB		(Batch_size, 14, 14, 64)
MaxPoolB	2x2, 2x2	(Batch_size, 14, 14, 64)
Fullyconnc	1024	(Batch_size, 3136)
Softmax	10	(Batch_size, 10)

python train_softmax_clean.py
--train_csv ./fashionmnist/training2.csv
--batch_size 100
--buffer_size 15000
--lr 0.0001
--log_dir ./log
--model_dir ./model
--nrof_epochs 5

To Do

• Add Summaries and Plots
• Save checkpoint
• Add validation loss and accuracy
• Export Model pb
• Early Stopping
• Add image summary of activation maps