NLStruct

Code used to produce experimental results for the paper "Deep Structured Prediction with Nonlinear Output Activations"

Welcome!

Requirements:

This code was developed using the following versions of the following libraries. I cannot guarantee that it will work with other versions:

• Python 3.5.2
• numpy 1.15.2
• PyTorch 0.4.0
• torchvision 0.2.0
• Matplotlib 2.2.2 (though you can just comment these parts out if necessary)
• Pillow 5.1.0 (for experiments requiring images)
• scikit-image
• dill

Additionally, running this code requires compiling a python extension written in C++ - this will require Python dev libraries as well as a C++ compiler. To compile this extension, enter the 'deepstruct/fastmp' directory and execute: python setup.py build

Structure

The python scripts used to run the experiments are found in the experiments folder, while shared model code is in the deepstruct folder.

Data

The data used for the word recognition experiments can be found in the data folder. The other datasets can be downloaded from various locations:

• Weizmann Horse Database
• MIRFLICKR25k
• Bibtex/Bookmarks

Short disclaimer

This repo consists of code that was taken from a larger research codebase. I attempted to prune everything so that only the relevant code remains; if something seems to be missing, just let me know and I'll make sure to track it down and add it back in. Thanks for taking a look!