https://github.com/deshanadesai/vqa-dataaugmentation

Data Augmentation for Visual Question Answering
https://github.com/deshanadesai/vqa-dataaugmentation

question-answering visual

Last synced: 18 days ago
JSON representation

Data Augmentation for Visual Question Answering

• Host: GitHub
• URL: https://github.com/deshanadesai/vqa-dataaugmentation
• Owner: deshanadesai
• Created: 2018-03-31T21:10:53.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2022-12-08T00:52:07.000Z (about 2 years ago)
• Last Synced: 2024-04-21T17:05:42.343Z (9 months ago)
• Topics: question-answering, visual
• Language: Jupyter Notebook
• Homepage:
• Size: 1.74 MB
• Stars: 7
• Watchers: 3
• Forks: 1
• Open Issues: 13
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
### VQA Data Augmentation

Baseline: Show, Ask, Attend, and Answer: A Strong Baseline For Visual Question Answering 

(paper: https://arxiv.org/pdf/1704.03162.pdf) 

(code: https://github.com/Cyanogenoid/pytorch-vqa)

Datasets: VQA2.0 http://www.visualqa.org/download.html

Evaluation: https://github.com/GT-Vision-Lab/VQA

Experiments To Do:

To run today:

* Add converse substitution to Language Only augmentation 

* Multiple word substitutions.

* Do some paraphrasing for known question types.

* How many  / Color of  - question substitution with hypernym doubt.

* Language augment other methods.

* Change all augmentation methods to fit the same vocab.

* Filter conceptnet based on question repetition.

* Add all working methods together for data augmentation.

* Make custom test set for places 365? Places 365 has adjectives as well as scene understanding.

* Add augmentation on image based on wrong answer or image type?