Download the semantic parser to $SEMANTIC_PARSER

mkdir daquar
cd daquar
wget
http://datasets.d2.mpi-inf.mpg.de/mateusz14visual-turing/wholeset_human_894.zip

unzip wholeset_human_894

cd train

In the NIPS paper for every test image we use a knn approximation, so
that for every test image we train a separate semantic parser.
This is reflected in the 'train' folder, where subfolder corresponds to
a test image. For instance 'train/image1010' corresponds to the test
image1010. The content of every such subfolder must be replaced with the
semantic parser. Let's take image1010 as a running example.

cd image1010
cp -R $SEMANTIC_PARSER .

cp ../../run_dcs_train.bash .

bash run_dcs_train.bash
This script will start training the semantic parser.

Repeat the same procedure for every image folder 'image*'.

Once the models are trained copy the weights in
'train/image1010/state/execs/0.exec/params.x' into
'test/image1010/state/execs/trained_params'

cd test/image1010
cp ../../run_dcs_test.bash .
cp -R $SEMANTIC_PARSER .

bash run_dcs_test.bash
This script produces runs the semantic parser and produces a log
file 'test/image1010/state/execs/0.exec/log'.
The answers must be next distilled from the log file and given 
to the WUPS score script (which can be downloaded from
http://datasets.d2.mpi-inf.mpg.de/mateusz14visual-turing/calculate_wups.py).