From 18752661e7eb23b17b5b80807df7332e8afdee04 Mon Sep 17 00:00:00 2001 From: Kevin Shih Date: Sun, 6 Aug 2017 18:26:06 -0500 Subject: [PATCH] reformatted README --- README.md | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/README.md b/README.md index 0bc309b..17318f7 100644 --- a/README.md +++ b/README.md @@ -1,8 +1,8 @@ -REQUIREMENTS: +## REQUIREMENTS: MATLAB 2015a or newer (for native python support) MatConvNet beta19 http://www.vlfeat.org/matconvnet/ -SETUP INSTRUCTIONS: +## SETUP INSTRUCTIONS: 1. Create a data directory 2. Setup directory paths as specified in globals.m. Relevant tarballs specified in comments 3. Clone the VQA API into the data directory and setup annotations @@ -11,7 +11,7 @@ SETUP INSTRUCTIONS: 6. Create results directory to store model snapshots from training 7. Download and extract text feature caches to top level directory (wget http://xor.cs.illinois.edu/~kevin/wtl_cache_feats/word2vec_cache_utils.tar.gz) -MAIN FUNCTIONS: +## MAIN FUNCTIONS: word_and_vision_regions_inner_network.m : running this should initialize training. Results stored in opts.train.expDir word_and_vision_regions_inner_network_init.m: constructs the network mcqMaxMarginLossLayer.m: Loss layer implementation @@ -19,11 +19,11 @@ regionsProjectInnerLayer2.m: region selection layer implementation determiner_list.m: list of removed stopwords removed from questions globals.m: contains global paths to where cached features are stored. -VISUALIZATION EXAMPLE: +## VISUALIZATION EXAMPLE: run visualize_on_held_out.m to visualize results on the held-out set. The held out set comprises 10% of the training data from the train set. Our test model can be downloaded from: http://xor.cs.illinois.edu/~kevin/wtl_cache_feats/wtl_trainval_model.mat -DIRECTORIES: +## DIRECTORIES: word2vec_cache_utils: directory that holds caches of pre-processed question and answers utils: misc utility functions