Skip to content

Latest commit

 

History

History
39 lines (28 loc) · 1.82 KB

datasets.md

File metadata and controls

39 lines (28 loc) · 1.82 KB
Raw

Dataset Preparation

This folder contains quick guidelines about the proposed CRIPP-VQA dataset.

Instructions

As the video files are too large to manage on cloud platforms, we provide only annotations files and video features. However, to generate the videos from scratch, please follow the below guidelines. This will benefit those who need raw visual data.

Note: Although we provide annotations for all counterfactual actions, the CRIPP-VQA challenge does not utilize videos or annotations of counterfactual scenarios during training/testing. But feel free to use this script to play around with different experiments.

Download

  • Annotation files are available at link.
  • [Optional] Mask-RCNN features are also available at link.

Video Generation

Follow the below steps to generate the video from the i.i.d. annotation files:

# setup repo
git clone git@github.com:Maitreyapatel/CRIPP-VQA.git

# setup env
cd CRIPP-VQA/dataset
virtualenv venv
source ./venv/bin/activate
pip install -r requirements.txt

If there is an issue with TDW installation, please refer to the official docs at GitHub.

We assume that annotation files are extracted inside the CRIPP-VQA/dataset/annotations/. And the generated videos will be stored inside CRIPP-VQA/dataset/generated_videos.

source ./venv/bin/activate
sh ./image_generation.sh <folder-name-inside-annotation-folder>

Note: Feel free to modify the image_generation.sh script according to your use case. Having errors !? (look below)

Issues

For technical concerns please create the GitHub issues. A quick way to resolve any issues would be to reach out to the author at maitreya.patel@asu.edu.