We have released Neural Network Libraries 1.2.0! Pre-trained DeepLabV3+ model API has been added!

Spotlight

Add DeepLabV3+ model as a pretrained model API

DeepLabV3+ is a state-of-the-art model for semantic segmentation, and is now readily available with pre-trained model API!

Layers​

• Add CUDA implementation of OneHot function

• RandomFlip implementation in CUDA

Build​

• Fix build

• Includes cuda lib in wheel

Utilities

• Modified pycpp-api to multiple output functions

• Generate data for parameter variables.

• Fix unexpected Reshape operations

• Add an option to keep source device and dtype in NdArray.copy_from

• Skip allreduce if array is not updated

• Fix callback import error and dcm-specified function obtained by call…

• Remove DICOM format support from neural network libraries.

​

Examples

• An example of fine-tuning YOLOv2 model on smaller dataset

​

Bug Fix / Documentation

• [Fix] Glorot Initializer

• Fix unsafe usage of contextmanager in auto_forward & resolve pytest fail

• fix convolution segv caused by cudaEvent_t instance variables

• Generate random seed when seed==-1 in cudnn context

• Update installation guide

• Update nnabla converter support status doc.

• Remove unused import lines

• Remove unused import lines

• remove unused import lines

Notice for Dropping Support for Python 2

• The Python core team plans to deprecate Python 2 support in 2020.
  NNabla project is currently supporting both Python 2 and Python 3 in parallel, but we are planing to drop support for Python 2 eventually.
• Our current plan is as follows:
  Until December 31, 2019, NNabla will fully support both Python2 and Python3.
  After January 1, 2020, NNabla will support Python3 only and will cease to support Python 2.