/MAC-Network

Memory, Attention and Composition (MAC) Network for CLEVR implemented in Knet

Primary LanguageJupyter NotebookMIT LicenseMIT

Compostional Attention Networks for Machine Reasoning

Knet implementation of the paper "Compositional attention networks for machine reasoning." Hudson, Drew A., and Christopher D. Manning.

Running Demo

Open visualize.ipynb notebook with jupyter to see the demo.

Getting Data

You have two options for the data setup.

a) Raw Data

1-Download CLEVR dataset to data/ folder.

2-Process the CLEVR data:

julia trainsetup.jl data/CLEVR_v1.0

b) Processed Data

1-Download preprocessed data:

julia trainsetup.jl

Training

Below configuration can achieve %98.27 accuracy on CLEVR dataset

julia train.jl src/main.jl configs/config2.jl