- HW Multibatch for FC layers
- Multi-input network support
- Support different precision and format for input tensors
- Buffer pre-registration
- INT8 deconvolution
- Deconvolution optmization
- Support deconvolution with stride > 32
- INT8 group convolution
- Depthwise convolution optmization
- ReLU-N
- Machine Translation Layer (MTL)
Note: APIs are expected to change in DLA1.3.0
- Memory optimzations
- ONNX
- Sample application for accuracy
- Sample application for object detection