Caffe Framework for DesignWare EV Processors, v2019.09-RC1

This is release 2019.09-RC1 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2019.09, from Synopsys.

New Features in this Release:

• evconvert enhancement: Add 13 new C++ layers and several new parameters for existing layers to support more TensorFlow and ONNX models conversion, e.g. ShuffleNet, BiseNet, LEDNet etc.
• python3 support enhancement: Update customized python layers implementation and build & dependencies scripts to make both python2 and 3 work compatibly
• Provide the detailed environment build guideline for Ubuntu platform

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2019.09-ENG1

This is release 2019.09-eng1 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2019.09, from Synopsys.

New Features in this Release:

• Add support for Linux/Windows platform build using python 3.6
• Update Windows dependencies library file (protobuf updated to 3.7.1, which is compiled using caffe-builder)

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2019.06

This is release 2019.06 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2019.06, from Synopsys.

New Features in this Release:

• evconvert enhancement: 16 new C++ layers added for supporting the conversion of TensorFlow Ops, such as the Tensor Dimension expansion and squeeze, broadcasting support for several Eltwise Ops
• Object Detection support refinement: several new parameters added for PriorBox layer to support the models converted from TensorFlow object_detction model zoo
• Add support for Windows platform build using python 3.6

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2019.06-RC1

This is release 2019.06-RC1 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2019.06, from Synopsys.

New Features in this Release:

• evconvert enhancement: 3 new C++ layers added for supporting the conversion of TensorFlow Ops
• Object Detection support refinement: several new parameters added for PriorBox layer to support the models converted from TensorFlow object_detction model zoo
• Add support for Windows platform build using python 3.6

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2019.03-RC2

This is release 2019.03-RC2 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2019.03, from Synopsys.

New Features in this Release:

• evconvert enhancement: 15+ new C++/python layers added for supporting the conversion of TensorFlow Ops
• Mask RCNN: 10+ new C/C++ layers and customized python layers added to support the conversion from TensorFlow/Keras model
• LSTM: TensorFlow-styled forget_bias support provided
• Bug fix: innerproduct layer CUDA implementation transpose case

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2019.03-RC1

This is release 2019.03-RC1 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2019.03, from Synopsys.

New Features in this Release:

• Mask RCNN: 10+ new C/C++ layers and customized python layers added to support the conversion from TensorFlow/Keras model
• LSTM: TensorFlow-styled forget_bias support provided
• Bug fix: innerproduct layer CUDA implementation transpose case
• evconvert enhancement: more C++/python layers added for supporting the conversion of TensorFlow Ops

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2018.12

This is release 2018.12 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2018.12, from Synopsys.

New Features in this Release:

• Semi-annual update with BVLC Caffe, synchronized to its latest commit 99bd997
• Yolo v2 and v3: upsample_darknet_layer added, training support provided
• evconvert: added depthtospace_layer and several customized python layers to support TensorFlow model conversion
• evquantize: more scaling parameters and saturate operations added for CUDA implementation in several layers
• Code refinement for Windows platform support

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2018.12-RC2

This is release 2018.12-RC2 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2018.12, from Synopsys.

New Features in this Release:

• Semi-annual update with BVLC Caffe, synchronized to its latest commit 99bd997
• Yolo v2 and v3: upsample_darknet_layer added, training support provided
• evconvert: added depthtospace_layer and several customized python layers to support TensorFlow model conversion
• evquantize: more scaling parameters and saturate operations added for CUDA implementation in several layers
• Code refinement for Windows platform support

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2018.12-RC1

This is release 2018.12-RC1 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2018.12, from Synopsys.

New Features in this Release:

• Semi-annual update with BVLC Caffe, synchronized to its latest commit 99bd997
• Yolo v2 and v3: upsample_darknet_layer added, training support provided
• evconvert: added depthtospace_layer and several customized python layers to support TensorFlow model conversion
• evquantize: more scaling parameters and saturate operations added for CUDA implementation in several layers

Full details see: FEATURES.md

Caffe Framework for DesignWare EV Processors, v2018.09

This is release 2018.09 of Synopsys-caffe, a modified version of the popular Caffe Deep Learning framework adapted for use with DesignWare EV6x Processors.

Synopsys-caffe is meant to be used together with synopsys-caffe-models and the MetaWare EV Development Toolkit v2018.09, from Synopsys.

New Features in this Release:

• evconvert: Refinement for asymmetric padding (pad_l, pad_r, pad_t and pad_b) and other related features
• added several customized python layers to support TensorFlow model conversion
• Patches for evprune support
• Support for Yolo_v2 and Yolo_v3 testing and training features
• evquantize: (evgencnn Inference acceleration) related parameters added for CUDA implementation
• Support refinement for Flownet2 (Dispnet) patches

Full details see: FEATURES.md