Ground Truth Layouts for VQA/VQA 2.0 - Not reproducible

[goonmeet]

Hello!

I am trying to generate ground truth layouts for the TDIUC dataset (which is very similar to the VQA and VQA 2.0 datasets) and does not contain functional programs.

Based on the ReadMe, I used these scripts as a starting point: https://gist.github.com/ronghanghu/67aeb391f4839611d119c73eba53bc5f

As a sanity test, I attempted to generate the layouts for the VQA and VQA 2.0 datasets. However, the generated layouts did not match the GT layouts provided in the repo.

The mismatch between the generated layouts might be because of the output of the Lexical Parser.

The following command is provided in the parse.py script: java -mx150m -cp "$scriptdir/*:" edu.stanford.nlp.parser.lexparser.LexicalizedParser
-outputFormat "words,typedDependencies" -outputFormatOptions "stem,collapsedDependencies,includeTags"
-sentences newline
edu/stanford/nlp/models/lexparser/englishPCFG.ser.gz \

However, this command does not output a parse tree in the following format: (ROOT (SQ (VBP are) (NP (EX there)) (ADJP (VBG patients)) (. ?)))

Instead, the following command can output something similar to the parse tree format mentioned above: java -mx150m -cp "*" edu.stanford.nlp.parser.lexparser.LexicalizedParser
-outputFormat "oneline" -outputFormatOptions "stem,collapsedDependencies,includeTags"
-sentences newline edu/stanford/nlp/models/lexparser/englishPCFG.ser.gz questions.txt

Based on this analysis, a few questions:

1. Is there a specific version of Lexical Parser that was used?
2. What specific command generates the output needed to run the parser?
3. Is the intermediate output from the parser available anywhere?

Thanks in advance for your time and help! Greatly appreciated!