RUArt

This repository is the implementation of "RUArt: A Novel Text-Centered Solution for Text-Based Visual Question Answering".

This repository is based on and inspired by @microsoft's work . We sincerely thank for their sharing of the codes.

Directory structure

Download pytorch bert-base-uncased model from huggingface, then extract it into #root/source, where #root is the code root folder.

The directory structure is as follows:

• RUArt
  • conf~
    • model (Downlod the pretrained RUArt model to this folder.)
  • Models
  • Utils
  • source
    • data (Download the preprocessed training and test files and extract it into this folder.)
    • bert-base-uncased
      • bert_config.json
      • pytorch_model.bin
      • vocab.txt
  • conf
  • main_test.py
  • readme.md

Requirements

pip install

pip3 install -r requiresments.txt

Inference

cd #root
python main_test.py

The result of ST-VQA task3 test will be saved in conf~/model/submission.json