tueimage/crowd-multitask

crowd-multitask's Issues

Decide on format for storing experimental results

Run experiment with crowdsourced border & color

Extract automatic features (asymmetry, border and color)

Code for extraction of asymmetry, border and other features is already available in Matlab (ask Veronika), needs to be translated to Python so that we can reproduce the results

Run different network with crowdsourced & automatic features

Run experiment with automatic asymmetry, border, color

Reproduce figure from Elif's paper

Create a training/test functions where features can be specified

Create functions which allow switching to different features. In bad pseudocode:

function train_network(feature_type)

% feature_type is a string, like "none", "asymmetry", "border", etc.
% For "none", a classification only network is trained
% For any others, load the .csv file with the required feature values, and train a multi-task network.
% Save the trained model
end

Same for test_network(feature_type). Load model, test it and save the predictions (probabilities per image).

Create a new layout and add Jupyter notebooks describing functionality

The issue is done when at least: 1. all parts have the new layout, 2. for each part a Jupyter notebook with documentation is present, and 3. the code able to produce the results described in Elif’s report