| Read our WACV-2022 Paper (HTML) | Who is shilling this tech? |
🥂🥳 Coverage in the Feb 2022 issue of the RSIP-Vision magazine 📰
Table of Contents
Summary
For the past year, we have been investigating this murky yet fascinating tech called Saliency cropping that passively touches all our lives silently influencing the ways in which we encounter digital images on the world wide web. Sold under different monikers such as ‘AI assisted cropping’, ‘Smart cropping’ and ‘Content aware cropping’, it has come to be extensively used by all the major digital content serving platforms such as Twitter, Facebook, Google, Microsoft and Apple, that seek to deliver their image-laden content on a wide array of consumption devices of varying screen sizes, aspect ratios and viewing angles. While Twitter had to grapple with the downstream cropping biases experienced by its users in a very public way ultimately resulting in its removal, it continues to be deployed aggressively across the other platforms.
We created this website to catalogue all our investigative efforts spanning 15 months, 3 continents, 2 papers, and, well, 2 COVID diagnoses :( It captures much of our efforts in the period spanning Sep 2020 to Nov 2021 (See the Timeline) during which we collaborated with computer vision researchers who have worked in this area of research and ethicists alike (See Acknowledgements).
In order to ensure transparency, we are sharing the reviews provided by the anonymous reviewers at the CVPR-2021 workshop and WACV here. In the coming week(s), we will be sharing all the code and a whole slew of experimental results that did not make it to either of our two papers. It is our goal to eventually summarize all our findings into a single monologue. If you are a researcher with relevant expertise who’d like to join hands, feel free to contact us 🤗
Video-Poster
| Video | Poster |
 |
Slides
Timeline
| Date | Event |
| September 19, 2020 | 🔭Discovery on Twitter and first simple set of experiments 🔬 |
| September 19, 2020 | ✍Created the @cropping_bias Twitter account to crowd-source the instances of offensive crops that informed our experiments |
| September 21, 2020 | ✍First blogpost titled ‘On the Twitter cropping controversy: Critique, clarifications & comments’ |
| October 2, 2020 | ✍ Second blogpost on Scrutinizing Saliency-Based Image Cropping |
| October 9, 2020 | A 30 min discussion with 2 members of a ‘meta-team’ within Twitter |
| April 2, 2021 | 😍 ✍ First workshop paper submission to the BeyondFairCV workshop (as Submission 21) |
| May 23, 2021 | 😎 Acceptance notification (Original decision date was Apr 23, but got postponed on account of the pandemic) |
| June 21, 2021 | ✍ Submission of the camera-ready version of the paper to the organizers of the workshop |
| June 25, 2021 | 🙌 Paper presentation at the workshop via Discord |
| August 11, 2021 | ✍ Second paper submitted @ WACV-2022 (Round-2) |
| October 4, 2021 | Acceptance notification at WACV-2022 🙌 |
| January 6, 2022 | 😎 🎊 Paper + Poster presentation at WACV-2022 |
| February 6, 2022 | 😎🙌 WACV 2022 paper as part of the Feb issue of the RSIP vision magazine |
| August 1, 2022 | 👉 Updated the landscape of current real-world deployments |
Interactive-demos
List of publicly available interactive demos maintained by researchers (As of Aug 2022)
Acknowledgements
This work would not have been possible without the incredible support of the following scholars: AbuBakar Abid (Gradio), Ali Abdalla (Gradio), Alexander Kroner, Sang Han, Prof. Broderick Turner, Darrell Owens, the organizers of the BeyondFairCV CVPR-2021 workshop, the attendees who gave us crucial feedback and all the volunteers who alerted us on various manifestations of cropping biases via our Cropping_bias Twitter handle and help with the hand-annotations in our experiments.