v0.4.0

• Mini-Caffe can be compiled on Windows, Linux, Android platform (OpenBLAS and protobuf needed)
• Add CUDA and CUDNN code back from original Caffe implement
• Remove all backward functions, Mini-Caffe now only provides Forward functions
• Optimize some layer implement (Since we don't need backward)
• Add C/Java/Python API
• Add examples to show how to integrate Mini-Caffe into other projects
• Refactoring code to tweak the API, remove all template, use float as default data type (Some API also changed)
• Add Profiler that helps analyse the network performance
• Add Layer Crafter in Python API for creating network prototxt

v0.3.0

• Mini-Caffe now only depends on OpenBLAS and protobuf.
• Remove all data layer including MemoryDataLayer, use InputLayer instead
• Remove all gpucode
• Replace glog with dmlc/logging

[email protected]

update to BVLC/caffe@f28f5ae