Please see below the instructions on downloading the raw images of CIRR.


Recently, we have received emails from a number of groups stating that the NLVR team is not responding to their requests.


Please know that we only provide this alternative download option after we have confirmed with you that the NLVR2 Google form “requesting raw images” has been submitted — therefore, you agree to their terms of service.


Note that we only provide the relevant images used in our dataset, which is not the entire NLVR2 image set.


Please download the images (zipped) at https://drive.google.com/file/d/1iFyNFHxsvvWFhnH9xDnhIgVJ49_Ngxrj/view?usp=sharing


After downloading, please check the file SHA1 and compare it with the value recorded in https://drive.google.com/file/d/1q_z6eivTEnyp-4dGQISAhivrlUfc6KpF/view?usp=sharing.


The image folders are structured exactly as demonstrated in our GitHub repository’s README page.



As a final remark, if you are the author/s of the NLVR2 dataset and reading this page, I would appreciate it if you could contact me personally (contact details are available in my home page) so that we could discuss the continuation of the distribution of the raw images for both our datasets.



This page is last modified on ‘May 14 2024’.