v0.3.3-p0

• Once more reduce memory usage when loading a model

v0.3.2-p0

• Further reduce memory usage when loading a model

v0.3.1-p0

• Reduce memory usage when loading a model from file

v0.3.0-p0

• add support for variably shaped input tensors
• new layers supported: depthwise_conv_2d_layer, PReLU, relu6, average_layer, multiply_layer and subtract_layer
• improve error handling and messages
• fixed some minor bugs
• add support for MobileNet
• compatibility to Keras 2.2.0

v0.2.3-p0

Conan support

v0.2.2-p0

Hunter support

v0.2.1-p0

Hunter support, CMake modernization

v0.2-p0

• Im2col convolution during prediction became faster, since some calculations now already happen already when loading a model.
• Support for non-im2col convolution was dropped.

v0.1-p0

add remarks about binary size and 32-bit compatibility