Skip to content

Caffe Framework for DesignWare EV Processors, S-2021.12
Compare
Choose a tag to compare
@vera121 vera121 released this 10 Jan 05:44
· 106 commits to master since this release
S-2021.12
e73a136

This is release S-2021.12 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV Family of Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit S-2021.12, from Synopsys.

New Features in this Release:

  • Ev support enhancement
    • Refined SpatialBatchingPooling and MatMul layer implementation
    • Added round_mode (derived from origin bvlc Caffe) in Pooling Parameter
  • Custom Quantized model support
    • Added per_channel_scale_output feature for Conv layer
    • Fixed depthwise Conv quantize rounding computation issue
  • Added support for sparse submanifold implementation in Conv layer
  • Uploaded CentOS build script with full environment setup; Updated Ubuntu build script with newer packages dependency support
  • Small updates in python scripts to avoid deprecated API warnings

Full details see: FEATURES.md

Assets 2
Loading