- Deploy on either SC3 or SE3;
- Replace some third-party libs to head only ones.
- Use new apis of BMNNSDK v1.1.3;
- Add samples for MXNnet and PyTorch models;
- Update documents by sphinx and abandon cpp document by doxygen;
- Deploy on SC3;
- Integrate video decoder, image processing, NN inference, postprocessing;
- Provide both C++ and Python APIs, write inference application with short code;
- Automatically manage memory of tensors;
- Automatically split models from Tensorflow and Mxnet for BM1682;
- Provide algokit(a python algorithm toolkit) to deploy deep learning applications;