• I am very interested in painting and I have always been wondering whether art experts were using only the painting to determine the style of the art piece or additional information such as the date, the name of the painter and so on.
• To understand it, I have trained a neural network using only the paintings to classify paitings among the following categories:
  • Art Nouveau
  • Baroque
  • Expressionism
  • Impressionism
  • Post-Impressionism
  • Rococo
  • Romanticism
  • Surrealism
  • Symbolism

• Performance on test set is 49%. Validation and training are Train 67.1% and 51%
• Optimal error rate is 30% because some categories are very similar such as impressionism and post impressionism, without additional information than the pixels like painter name and date a person knowledgeable about art would reach 70% accuracy on this dataset (calculated on 50 examples from the validation set)
• Final neural network model used transfer learning based on a vgg19 with a fully connected layer containing 12595 hidden units trained on 1600 examples implemented with Pytorch
• I used many techniques to reduce avoidable biais and variance
  • avoidabale biais: increase the model size (additional layer in the classifier, additional neurons in the layer, tested other architecture with resnet50), tested different regularization parameters
  • variance: added more training data and early stopping

• painting_classifier.ipynb: jupyter notebook containing the training, prediction, error analysis of the DL model

• checkpoint5.pth: the saved version of the model (download it here: https://drive.google.com/open?id=1C75piZ_YxpFzOQyhif_ERpaWov5PC4gd and save it in the same folder as the app/)

• app/: contains the python script for an application in the terminal that predicts the painting style of a painting

• Upload an image of a painting in app/example for e.g. 100152.jpg
• Basic usage: run python predict.py --image example/100152.jpg
• Options:
  • Return top 5 most likely classes: python predict.py --image example/100152.jpg --top_k 5

absl-py 0.4.1 annoy 1.12.0 args 0.1.0 astor 0.7.1 atari-py 0.1.1 backcall 0.1.0 bleach 2.1.4 Box2D-kengz 2.3.3 cachetools 2.1.0 certifi 2018.8.13 chardet 3.0.4 click 6.7 clint 0.5.1 cloudpickle 0.5.3 cupy 4.3.0 cupy-cuda91 4.3.0 cycler 0.10.0 cymem 1.31.2 Cython 0.28.5 cytoolz 0.9.0.1 dask 0.18.2 decorator 4.3.0 dill 0.2.8.2 dlib 19.15.0 entrypoints 0.2.3 fastrlock 0.3 Flask 1.0.2 floyd-cli 0.11.17 ftfy 4.4.3 future 0.16.0 gast 0.2.0 grpcio 1.14.2 gym 0.10.5 gym-retro 0.5.6 h5py 2.8.0 html5lib 1.0.1 idna 2.7 ijson 2.3 incremental 17.5.0 intel-openmp 2019.0 ipykernel 4.8.2 ipython 6.5.0 ipython-genutils 0.2.0 ipywidgets 7.4.0 itsdangerous 0.24 jedi 0.12.1 Jinja2 2.10 jsonschema 2.6.0 jupyter 1.0.0 jupyter-client 5.2.3 jupyter-console 5.2.0 jupyter-core 4.4.0 jupyterlab 0.33.8 jupyterlab-launcher 0.11.2 kaggle 1.4.6 Keras 2.2.0 Keras-Applications 1.0.2 Keras-Preprocessing 1.0.1 kiwisolver 1.0.1 Markdown 2.6.11 MarkupSafe 1.0 marshmallow 2.15.4 matplotlib 2.2.3 menpo 0.8.1 mistune 0.8.3 mkl 2019.0 mkl-devel 2018.0.3 mkl-include 2018.0.3 mpmath 1.0.0 msgpack 0.5.6 msgpack-numpy 0.4.3.1 murmurhash 0.28.0 nbconvert 5.3.1 nbformat 4.4.0 networkx 2.1 nltk 3.3 notebook 5.6.0 numpy 1.15.1 opencv-contrib-python 3.4.0.12 pandas 0.23.4 pandocfilters 1.4.2 parso 0.3.1 path.py 11.0.1 pathlib2 2.3.2 pexpect 4.6.0 pickleshare 0.7.4 Pillow 5.2.0 pip 10.0.1 plac 0.9.6 plotly 3.1.1 preshed 1.0.1 prometheus-client 0.3.1 prompt-toolkit 1.0.15 protobuf 3.6.1 ptyprocess 0.6.0 pydot 1.2.4 pyemd 0.5.1 pyglet 1.3.2 Pygments 2.2.0 pynvrtc 9.2 PyOpenGL 3.1.0 PyOpenGL-accelerate 3.1.0 pyparsing 2.2.0 Pyphen 0.9.4 python-dateutil 2.7.3 python-Levenshtein 0.12.0 pytz 2018.5 PyWavelets 0.5.2 PyYAML 3.13 pyzmq 17.1.2 qtconsole 4.4.1 raven 6.9.0 regex 2017.4.5 requests 2.19.1 requests-toolbelt 0.8.0 retrowrapper 0.2.1 retrying 1.3.3 scikit-image 0.14.0 scikit-learn 0.19.2 scikit-umfpack 0.3.1 scipy 1.1.0 seaborn 0.9.0 Send2Trash 1.5.0 setuptools 39.1.0 simplegeneric 0.8.1 six 1.11.0 sklearn 0.0 spacy 2.0.12 sympy 1.2 tabulate 0.8.2 tensorboard 1.9.0 tensorboardX 1.2 tensorflow 1.9.0 termcolor 1.1.0 terminado 0.8.1 testpath 0.3.1 textacy 0.6.2 tflearn 0.3.2 thinc 6.10.3 toolz 0.9.0 torch 0.5.0a0+a24163a torchtext 0.3.1 torchvision 0.2.1 tornado 5.1 tqdm 4.25.0 traitlets 4.3.2 typing 3.6.6 ujson 1.35 Unidecode 1.0.22 urllib3 1.22 uWSGI 2.0.17 virtualenv 16.0.0 wcwidth 0.1.7 webencodings 0.5.1 Werkzeug 0.14.1 wheel 0.31.1 widgetsnbextension 3.4.0 wrapt 1.10.11 xgboost 0.80 zmq 0.0.0

• Dataset is from the kaggle competition painter by numbers and corresponds to the train_1.zip (https://www.kaggle.com/c/painter-by-numbers/data)
• Capstone project of Udacity DS Nanodegree program