0.3.1: Merge pull request #42 from Tensor46/develop

Layers

• DropBlock

• bug fix

v0.3: Merge pull request #39 from Tensor46/develop

New Modules:
Architectures -

• AnatomyNet
• UNET
• PointNet
• ResNeXt
• Squeeze and Excitation ResNet
• Squeeze and Excitation ResNeXt
• DenseNet
• Neural Decision Trees
• PGGAN
  Essentials -
• CudaModel (regularize_weights, clip weights)
• DataSets (mnist/fashionmnist/cifar10/cifar100)
• MakeModel (BaseModel, SaveModel and LoadModel moved to MakeModel)
• MakeGIF
• VisPlots (show_images, show_weights and histograms)
  Layers -
• Activations (relu/relu6/lklu/elu/prelu/tanh/sigm/maxo/rmxo/swish)
• Normalizations (batch/group/instance/layer/pixelwise)
• Linear (with all activations and flexible out_shape)
• Convolution (Convolution/ConvolutionTranspose + Normalization + Activation)
• LossFunctions (+ LCML, Center loss, LMGM loss, TripletLoss, DiceLoss)
• ConvolutionalSAE (Base block for stacked auto encoder)
  utils -
• corr_1d (correlation)
• xcorr_1d (cross correlation)
• DoH (determinant of Hessian)
• HessianBlob (Aggregates determinant of Hessian with various width)
• GaussianKernel
• DoG (Difference of Gaussian)
• DoGBlob (Accumulates DoG's at different scales)
• roc (reciever operating curve)
• ImageNetNorm (ImageNet normalization)

Deprecations:

• CIFAR10
• MNIST
• ConvolutionTranspose (merged to Convolution with transpose argument)