upgrade Tensorflow to 2.4.1

• cuda 11.0.3
• cudnn 8
• ubuntu 20.04
• python 3.8
• tensorflow-gpu 2.4.1
• pytorch 1.8.1

add py3 env; update frameworks

• Dockerfile.gpu is now split to Dockerfile.py3 and Dockerfile.py2
• upgrade tf to v1.11.0
• upgrade PyTorch to v0.4.1

add Caffe2

to test instance segmentation on https://github.com/facebookresearch/Detectron

upgrade tf, add torchvision

• upgrade tensorflow to v1.9-rc1
• add torchvision v0.2.1

cuda9.0 + cudnn7

Frameworks upgrade

• Tensorflow: 1.8.0-rc1
• nvidia caffe: 0.17
• Pytorch: v0.4.0

Other changes

• Pytorch installation is changed from precomputed library to compiling from source
  • pros: for keeping up with latest features
  • cons: doubles the image size as well; however, storage is cheap

tensorflow-pytorch-caffe

major update, cleanup lots of unused dependencies

Environment

• ubuntu 16.04
• cuda 8.0
• cudnn 6

DL frameworks included

• Tensorflow 1.4.1 (built from source)
  • Tensorboard 0.4.0
  • Tensorflow Profiler
  • Google pprof
• nvidia-Caffe 0.16
• pytorch 0.2.0

Other

• bashrc and vimrc are now setup automatically

various framework updates

cudnn 5->6
nvCaffe 0.15->0.16
tensorflow 1.0->1.3.0
pytorch 0.2.0
digits 5.0->6.0

Upgrade Frameworks to latest version

• Upgrade CUDA to 8.0.44
• Upgrade TensorFlow to 1.0.1-cp27
• Upgrade Torch to latest version
• Upgrade Theano to rel-0.9.0rc4
• Upgrade Keras to 2.0.1

Create .digits and add t-SNE to .gpu

Add t-SNE

https://github.com/oreillymedia/t-SNE-tutorial

Update theano to bleeding-edge

Use bleeding-edge theano

Solve https://github.com/Theano/Theano/issues/4759 on 0.8.2