Can't obtain Table III results with Li's 3D CNN network

[maudBqd]

Hi, I am testing Li's 3D CNN network on IndianPines dataset. In table III in disjoint mode, you have an accuracy of 75%. I can't reach more than 65%.

• Are the default parameters of the network (lr = 0.01, epoch = 200, patch_size = 5, momentum=0.9, weight_decay = 5e-4, n_planes = 16) the ones you have used ? In particular, in the original paper, n_planes = 2
• Did you just apply flip_augmentation or also mixture_augmentation or radiation_augmentation ? Even with those it doesn't work.
• The other parameters I used : training_sample = 0.9, default batch_size, class_balancing=True

Thank you :)

[nshaud]

I'll have to check which hyperparameters I used for this experiment.