icdar_vqa

Implementation in Pytorch for ICDAR challege, 2019. Paper -- https://arxiv.org/abs/1905.13648 Results -- https://rrc.cvc.uab.es/?ch=11&com=evaluation&task=1

The goal of this repository is to provide an implementation for all the 3 tasks of this challenge namely strongly contexualized, weakly contextualized and open dictionary.

Prerequisites

• python 3.6+
• numpy
• pytorch 0.4
• tqdm
• nltk
• pandas

Data

• Relesed by challenge organizors.
• COCO 36 features pretrained resnet model
• GloVe pretrained Wikipedia+Gigaword word embedding

Steps to use this repository

• First, download the data from challenge website https://rrc.cvc.uab.es/?ch=11&com=downloads

• Secondly, the following command is used to prepare data for training.

python prepro.py

-Once the data is ready for training, the training can be started using this command.

bash train.sh

Notes

-The results are reported in https://arxiv.org/abs/1905.13648. -We got 3rd position in 1st task.

• Some of preproc.py and utils.py are based on this repo

Resources

• The paper.
• Their CVPR Workshop slides.