Docker for ML/DL

Template for machine learning and deep learning projects.

Requirements

• Nvidia graphics card with CUDA support http://www.geforce.com/hardware/technology/cuda/
• Nvidia driver http://www.geforce.com/drivers
• Docker https://www.docker.com/get-docker
• Docker Compose https://docs.docker.com/compose/install/
• Nvidia Modprobe sudo apt-get install nvidia-modprobe

Installation

Run:

• install-nvidia-docker.sh
• install-nvidia-docker-compose.sh

Post-installation

Check docker volume ls and run nvidia-docker run nvidia/cuda nvidia-smi if driver volume is empty

---

Base images

8.0-cudnn5-devel-ubuntu16.04

Tensorflow-gpu

• rlxs/tensorflow:0.12.1-gpu-py2
• rlxs/tensorflow:0.12.1-gpu-py3
• rlxs/tensorflow:1.1.0-gpu-py2
• rlxs/tensorflow:1.1.0-gpu-py3

OpenCV 3.2.0 with CUDA support + Tensorflow-gpu 1.1.0 + Python2.7/3.5

• rlxs/opencv-cuda-tf

OpenCV 2.4.9.1 + Tensorflow-gpu 1.1.0 + Python2.7 (Jupyter/matplotlib/numpy/scipy)

• rlxs/tf-cv2

Torch 0.1.12_2 + Python2.7

• rlxs/cuda-torch:py2

---

Create images and test gpu support

Test GPU support in Tensorflow

tensorflow_gpu_test.sh

---

Docker OpenCV GUI test and webcam access

opencv_test.sh

---

Tensorbox https://github.com/TensorBox/TensorBox

• Run Download data to download video datasets for training

• Start tensorboard and tensorbox

• Test tensorbox on own dataset

---