Anaconda3, Jupyter Notebook, OpenCV3, TensorFlow and Keras2 for Deep Learning

Tag	Size / Layers	CUDA Toolkit (Linux x86_64 / Windows x86_64)	CUDNN
latest, cpu		-	-
10.0-cudnn7		CUDA 10.0 ( >= 410.48 / 411.31 )	7
9.2-cudnn7		CUDA 9.2 ( >= 396.26 / 397.44 )	7
9.1-cudnn7		CUDA 9.1 ( >= 387.26 / 388.19 )	7
9.0-cudnn7		CUDA 9.0 ( >= 384.81 / 385.54 )	7
8.0-cudnn6		CUDA 8.0 ( >= 375.26 / 376.51 )	6

Tag	Size / Layers	Base	Tensorflow
tf-cpu, tf		cpu	1.12.0
tf-10.0-cudnn7		10.0-cudnn7	1.12.0
tf-9.2-cudnn7		9.2-cudnn7	1.12.0
tf-9.1-cudnn7		9.1-cudnn7	1.7.0
tf-9.0-cudnn7		9.0-cudnn7	1.7.0
tf-8.0-cudnn6		8.0-cudnn6	1.4.1

Tag	Size / Layers	Base	Tensorflow	Keras
keras-cpu, keras		tf-cpu	1.12.0	2.2.4
keras-10.0-cudnn7		tf-10.0-cudnn7	1.12.0	2.2.4
keras-9.2-cudnn7		tf-9.2-cudnn7	1.12.0	2.2.4
keras-9.1-cudnn7		tf-9.1-cudnn7	1.7.0	2.1.5
keras-9.0-cudnn7		tf-9.0-cudnn7	1.7.0	2.1.5
keras-8.0-cudnn6		tf-8.0-cudnn6	1.4.1	2.1.3

Tag	Size / Layers	Base	Pytorch
pytorch-cpu, pytorch		cpu	1.0.0
pytorch-10.0-cudnn7		10.0-cudnn7	1.0.0
pytorch-9.2-cudnn7		9.2-cudnn7	1.0.0
pytorch-9.1-cudnn7		9.1-cudnn7	1.0.0
pytorch-9.0-cudnn7		9.0-cudnn7	1.0.0
pytorch-8.0-cudnn6		8.0-cudnn6	1.0.0

Tag	Size / Layers	Base	Mxnet
mxnet-cpu, mxnet		cpu	mxnet
mxnet-10.0-cudnn7		10.0-cudnn7	mxnet-cu100
mxnet-9.2-cudnn7		9.2-cudnn7	mxnet-cu92
mxnet-9.1-cudnn7		9.1-cudnn7	mxnet-cu91
mxnet-9.0-cudnn7		9.0-cudnn7	mxnet-cu90
mxnet-8.0-cudnn6		8.0-cudnn6	mxnet-cu80

1. Run with docker (image: okwrtdsh/anaconda3:keras-cpu)

$ docker run -v $(pwd):/src/notebooks -p 8888:8888 -td okwrtdsh/anaconda3:keras-cpu

2. Open http://localhost:8888 in web browser

1. Run with nvidia-docker (image: okwrtdsh/anaconda3:keras-10.0-cudnn7)

$ nvidia-docker run -v $(pwd):/src/notebooks -p 8888:8888 -td okwrtdsh/anaconda3:keras-10.0-cudnn7

2. Open http://localhost:8888 in web browser

1. docker-compose.yml (image: okwrtdsh/anaconda3:keras-cpu)

version: '3'
services:
  jupyter:
    image: okwrtdsh/anaconda3:keras-cpu
    ports:
      - '8888:8888'
    volumes:
      - ./notebooks:/src/notebooks

2. Run with docker-compose

$ docker-compose up -d

3. Open http://localhost:8888 in web browser

1. docker-compose.yml (image: okwrtdsh/anaconda3:keras-10.0-cudnn7)

version: '3'
services:
  jupyter:
    image: okwrtdsh/anaconda3:keras-10.0-cudnn7
    ports:
      - '8888:8888'
    volumes:
      - ./notebooks:/src/notebooks

2. Run with nvidia-docker

# Run with nvidia-docker-compose (nvidia-docker v1)
$ nvidia-docker-compose up -d
# Run with docker-compose (nvidia-docker v2)
$ docker-compose up -d

3. Open http://localhost:8888 in web browser

Please note that PyTorch uses shared memory to share data between processes, so if torch multiprocessing is used (e.g. for multithreaded data loaders) the default shared memory segment size that container runs with is not enough, and you should increase shared memory size either with --ipc=host or --shm-size command line options to nvidia-docker run.

$ nvidia-docker run --ipc=host -v $(pwd):/src/notebooks -p 8888:8888 -td okwrtdsh/anaconda3:pytorch-10.0-cudnn7

version: '3'
services:
  jupyter:
    image: okwrtdsh/anaconda3:pytorch-10.0-cudnn7
    ipc: host
    ports:
      - '8888:8888'
    volumes:
      - ./notebooks:/src/notebooks

# Run with nvidia-docker-compose (nvidia-docker v1)
$ nvidia-docker-compose up -d
# Run with docker-compose (nvidia-docker v2)
$ docker-compose up -d