1.0.md

ML Suite v1.0 Release Notes

This release is the first push of Xilinx's ML Suite to Github.
Releasing to github will enable rapid release cycles, a smaller release footprint, and open source contribution.

Features

• FPGA Accelerated Image Classification, support for many networks
• FPGA Accelerated Object Detection, YOLOv2 support
• Python API for deploying inference to FPGA
• Precompiled xfDNN Library
• Support for xDNNv2

Framework Support and Layers

• Caffe: 1.0.0
• Tensorflow: 1.7
• Supported Layers
  • Convolution
  • ReLU - supported following Convolution / Eltwise Layers
  • Pooling
  • Deconvolution
  • Concat
  • Eltwise
  • BatchNorm
  • Scale
  • Slice
  • Layers supported in CPU:
  • InnerProduct
  • Softmax

Known Issues

• Batch Normalization layers not supported by the quantizer
• Local Response Normalization layers not supported
• Hardware solution for average pool causes some accuracy loss, to be fixed in a future release
• Standard ReLU is the only supported non-linearity (Leaky ReLU Networks must be modified/retrained)