Exclusive: Deepfake pioneer to lose two papers after misconduct finding of faked data

Hao Li

Two papers coauthored by a computer scientist whose work on visual effects has been credited in big-name Hollywood movies will soon be retracted after a publisher’s investigation found falsification of data in the articles. 

Retraction Watch has also learned that the University of Southern California (USC) found that Hao Li “falsely presented his research” in the two publications while he was a professor there. The articles, both published in journals of the Association for Computing Machinery (ACM), describe a system to create a 3D digital avatar head from a person’s photo using artificial intelligence. 

Li co-founded and is CEO of Pinscreen, a startup which is commercializing that technology. On its website, Pinscreen touts its products as “the most advanced AI-driven versatile avatars.” Besides personalized avatars for use in virtual or augmented reality systems, Pinscreen offers the ability to replace a person’s face in videos, creating what’s known as “deepfakes.” 

MIT Technology Review named Li an “innovator under 35” in 2013, and in a 2019 profile the magazine called him “a pioneer of digital fakery.” The article also described Li’s work to create an onscreen image of the actor Paul Walker for the movie Furious 7, after Walker died halfway through filming. Li’s other visual effects credits include Blade Runner 2049 and The Hobbit: Battle of the Five Armies.

Both of the articles to be retracted date from 2017. “Avatar digitization from a single image for real-time rendering” was published in ACM Transactions on Graphics, and “Pinscreen: creating performance-driven avatars in seconds” accompanied a presentation at the ACM computer graphics conference SIGGRAPH 2017 Real Time Live!

In the presentation, which was uploaded to YouTube (beginning at 31:15), Li and several Pinscreen employees take the stage to explain and demonstrate the company’s technology. 

“What we’re going to show you later is how to build a high-quality 3D avatar from a single image, fully rigged, animatable, and we’re going to show you how to bring them to life using a single webcam,” Li said. 

To demonstrate, Iman Sadeghi, then Pinscreen’s vice president of engineering, took a picture of himself with a webcam, and a few seconds later, a 3D image of his head appeared on the computer screen, animated and changing its facial expressions. Sadeghi then proceeded to click on photos of other people, and their avatar heads appeared on the screen, apparently generated in real time. 

But according to a lawsuit Sadeghi filed later, and USC’s investigation, Pinscreen’s technology at the time of the presentation could not perform as it appeared to do. At the request of the university’s investigation committee, a consultant reviewed the computer code Sadeghi used to present and found that all of the avatars in the demonstration had been built beforehand. The model of Sadeghi’s head would appear no matter what picture the webcam took as an input, according to the consultant.  

Sadeghi sued Pinscreen in 2018 (as reported at the time by the LA Times and USC Annenberg Media), alleging Li had fired him for raising concerns about the company misrepresenting its technology in the SIGGRAPH Real Time Live! presentation and other venues. The suit is ongoing in Los Angeles Superior Court. 

After filing his suit, Sadeghi reported the falsification to ACM and USC, where Li was a professor in the Viterbi School of Engineering and director of the Institute for Creative Technologies Vision and Graphics Lab. USC began looking into the allegations. 

According to a version of USC’s investigation report posted on Sadeghi’s website, the investigation committee concluded that Li: 

Knowingly and intentionally presented a falsified demonstration of his software at the SIGGRAPH Real-Time-Live show on August 1, 2017 with the intention to mislead the audience into believing that they were viewing a real-time demonstration of the automatic avatar-generating software that he and his team claimed to have developed, when in fact, Dr. Li and his team presented pre-programmed, manually produced avatar generation.

A spokesperson for the engineering school has not responded to our request for comment on the report. 

Li’s research was funded by two federal grants. One, on which Li was the principal investigator, was from the Office of Naval Research. The title of the project in the attached grant matches the title of the Young Investigator Award that Li won to develop realistic avatars for virtual training systems for the Navy. USC’s announcement of the award in 2018 stated that it was worth $600,000 over three years. 

Li left USC in 2020, and, according to his website, spent the next two years as a distinguished fellow at the University of California, Berkeley. He is now an associate professor at Mohamed Bin Zayed University of Artificial Intelligence in Abu Dhabi. He did not respond to our request for comment. 

In responses to USC’s inquiry report, Li denied that he misrepresented Pinscreen’s technology and said that preloading the avatars for the presentation was allowed in case of connectivity issues. His response also references letters from SIGGRAPH organizers that he said support his claims. He wrote: 

Let me be very clear: there was absolutely NO fabrication and/or falsification from either our teams at USC or Pinscreen at any point in time. Nor did I or anyone associated with me mislead the public or the scientific community. 

ACM’s investigation concluded last year, and did not vindicate Li’s defense that the conference organizers had allowed the preloaded avatars as presented. According to a decision letter dated Sept. 14, 2022 and posted online to Sadeghi’s website, the ACM Ethics and Plagiarism Committee of the ACM Publications Board investigated complaints about the work and found the authors “guilty” of violating the organization’s policy on plagiarism, misrepresentation, and falsification: 

Specifically, the Committee found that the presentation at SIGGRAPH 2017 Real Time Live! falsely presented pre-generated images as dynamically created. This fact should have been disclosed to the audience, but was not. This data falsification extended to the ACM Transaction on Graphics article. 

The committee decided that the violation was “severe,” and imposed penalties including retraction of the articles and a five-year ban on submitting work to or serving in any official role for ACM publications. 

Scott Delman, ACM’s director of publications who signed the letter, told us the letter was “confidential” and “not intended to be made public.” 

Sadeghi also posted a letter he sent to ACM in response, asking for the organization to reconsider the penalties for himself. In his letter, dated October 14, Sadeghi wrote that he had reported the violations to the group, and to USC, and that he had “actively participated and provided information during both investigations and preserved key evidence.” 

He continued: 

As the CEO of Pinscreen and my direct superior in the company, Li exploited his power differential to force me into participating in his fraud. Li depended on that differential as well as my obligations under the Confidentiality Contract to keep his coercive, abusive, and fraudulent practices from being reported.

My involvement in Pinscreen’s fraudulent publications and presentations is an untoward consequence of Li’s coercive and abusive practices. The repeated whistleblowing and objections I presented to Li resulted in the retaliatory termination of my employment immediately after Pinscreen’s public deception at ACM SIGGRAPH Real-Time Live 2017. According to the policy, ACM should investigate this coercive and abusive situation and seek to remediate the untoward consequences and the retaliatory actions. 

Holding the perpetrator and the victim of coercion equally accountable for fraud simply because both are listed as the authors on the fraudulent publications is against the spirit of the Code.

Like Retraction Watch? You can make a tax-deductible contribution to support our work, follow us on Twitter, like us on Facebook, add us to your RSS reader, or subscribe to our daily digest. If you find a retraction that’s not in our database, you can let us know here. For comments or feedback, email us at [email protected].

80 thoughts on “Exclusive: Deepfake pioneer to lose two papers after misconduct finding of faked data”

      1. Well. Hao Li and Pinscreen have given up on the idea of automatically simulating the hair shapes and now they only do face swaps which has been doable for many years and before 2017.

        The fraudulent component of their publications (the hair shapes) still doesn’t work on their app after these many years.

        All his publications must be retracted.

        See FAC Paragraph 175:

        Li stated, on June 29, 2017, that he was “really worried that nothing would work” by the RTL rehearsal and that Pinscreen would have to do “some shitty cheating again.” (Exhibit E5)

          1. I agree that Pinscreen’s app results are nothing close to the quality of the now-shown-to-be fabricated results in the publications.

      1. Eye opening investigation. After USC asked to examine his USC computer and hard drives, Hao Li deleted everything. It is hard to believe that he is still in academia.

  1. Well done Retraction Watch for the full coverage of Hao Li’s and Pinscreen’s Scientific Misconduct at ACM SIGGRAPH and at USC.

    MBZUAI (Mohamed bin Zayed University of Artificial Intelligence) and its chair Eric Xing should know better than to keep Hao Li in the list of its faculty members.

    Not only MBZUAI has not addressed this research compliance issue so far, it has actively attempted to cover up the misconduct investigations:

    http://sadeghi.com/MBZUAI-CoverUp

    Everyone, please urge MBZUAI to top covering up and come clean.

    #Science #ScientificIntegrity #NotFraud #NoCoverUp #MBZUAI

  2. Was MBZUAI so desperate to hire Hao Li after USC caught him cheating? How is this guy still in academia at all?

    1. One would hope that MBZUAI was not aware of the misconduct at the time of hiring. But now there is no excuses …

    1. Thank you Dr. Iman Sadeghi for standing up for scientific integrity. We need more whistleblowers like you to ensure long term scientific prosperity.

    1. Scott Delman, ACM’s director of publications who signed the letter, told us the letter was “confidential” and “not intended to be made public.”

      Which letter is this? Why can’t it be made public?

      Overall, great coverage. Glad to know the technology Li claimed in the acm papers are fake.

      1. I believe the “confidential” letter is the one dated Sept. 14, 2022 which is already public and is linked from the RW article.

    1. MBZUAI must make an official response to clarify its position about Hao Li’s research ethics.

      See: In preparation for SIGGRAPH RTL submission, due on April 4, 2017, Li wrote on multiple team threads, on March 27, 2017, that “the issue is that we don’t have time,” and that “even if we fake things there is no time,” and that for the hair reconstruction (i.e. hair shape estimation) “we probably have no choice but to cheat.” (Exhibits E3, E7)

      http://sadeghi.com/dr-iman-sadeghi-v-pinscreen-inc-et-al/#p115

    1. Not surprising. Hao Li’s students just had defend their PHD advisor because their own PHD degrees were on the line.

  3. To this date Pinscreen’s app still cannot automatically generate the hair geometry of Haley shown in their RTL submission in 2017. Indeed as Hao Li wrote himself wo his team it is a “shitty cheating” (see the link below). I also believe that in addition to scientific misconduct, these falsifications count as investment fraud.
    http://sadeghi.com/dr-iman-sadeghi-v-pinscreen-inc-et-al/#p175
    “However, as Pinscreen approached the RTL public presentation date of August 1, 2017, on information and belief, Li realized that Pinscreen would not be able to deliver on Li’s inflated claims put forth in the submission, months earlier on April 4, 2017, despite Pinscreen employees’ long hours and hard work. Li stated, on June 29, 2017, that he was ‘really worried that nothing would work’ by the RTL rehearsal and that Pinscreen would have to do ‘some shitty cheating again.’ “

    1. Yes Pinscreen’s app always produced low quality avatars and wrong hair geometry and their desperate defense has been that the app has an “optimized” version! Nonsense.

      1. Pinscreen app doesn’t produces high quality avatars and I uninstalled it right away. I tried it again to see if they have made improvements but the hair style, hair geometry, hair color, eye color were all wrong. The body that they have added is zombie like and the definition of uncanny valley.

  4. ACM SIGGRAPH needs to publicize such findings of scientific misconduct rather than keeping it “confidential”.
    Thank you Retraction Watch and Dr. Iman Sadeghi for shedding light on academic fraud.

    1. I completely agree. ACM must announce the findings in a newsletter with details with investigation. ACM members and SIGGRAPH paper submitters need truth.

  5. And what happens to all the students who received their PHD degrees based on this retracted publications? Two SIGGRAPH papers makes a huge difference in a PHD defense decision.

      1. Thank you for the link Richard. What are your thoughts on what is the right thing to do in these situations?

        1. I believe all doctorate degrees that were awarded at least partially based on these fraudulent publications must be revoked. Two Siggraph papers are a significant factor in awarding these degrees.

          1. PhD degrees (as well as any MSc degrees) that contained these fraudulent publications must all be revoked and the individuals must defend their non-fraudulent work.

            But then of course with a resume stained with a history of scientific misconduct it is hard to imagine they could successfully convince any defense committee.

  6. I think the most striking part of the article is about ACM effectively authenticating the investigation letter but then declaring that the letter was supposed to be “confidential”.

    The fact that there is no such label anywhere on the letter aside, why would the letter be “confidential” if the retractions will be public? Why would hide that the SIGGRAPH conference reviewers where unable to catch this systematic academic fraud? Why not requiring the code along the submission to begin with?

    ACM should itself put a spotlight on these findings to discourage others from submitting and presenting fabricated results.

    1. I expect ACM to make an announcement after the papers are officially retracted and removed from its digital library.

  7. Glad to see the truth finally come out. ACM should clarify all findings of misconduct contained in these publications in its newsletter. The lawsuit has evidence regarding fabricated hair shapes, hair colors, hair segmentations etc. Which one of these claims have been investigated and substantiated? SIGGRAPH community deserves to know the whole truth.

  8. Thank you Retraction Watch for the comprehensive and concise coverage.
    It is truly mind-boggling that Sadeghi had to spend half a million dollars in order to expose this academic fraud. He has my ultimate support and respect.

  9. What took ACM this long to investigate this misconduct which was fully documented in writing and source codes as of 2017?

    1. To me looks like the fact that there has been an active lawsuit has slowed down the process quite a bit. Most institutions in the US do anything they can to minimize their liability and USC and ACM seem to be no different.

  10. Do others also feel like this just the tip of the iceberg of scientific misconduct in computer science and machine learning?
    I hope this will encourage other whistleblowers to come out..

  11. A lot of interesting comments on this article. Unusual levels of engagement for a RW post.
    I wonder if these are actually different people, or if Dr. Sadeghi is creating sock puppet accounts to drum up support for his lawsuit?

      1. My immediate reaction was the same … that the “concerned commenter” is most likely Li or otherwise related to Pinscreen.

  12. I found this comment section from the link below and am impressed by the amount of informative comments and wanted to add my two cents:
    Sadeghi’s story has been consistent since he filed the lawsuit in 2018. Li’s and his students’ story has changed over time when first they claimed 100% of Sadeghi’s claims were false, then claimed Sadeghi’s evidence was photoshopped, then claimed Pinscreen cached the avatars due to potential connectivity issues, and now the full investigations.

  13. I’m happy to see the result of this long battle finally come to light.
    The USC investigation had to be subpoenad from the University and ACM investigation is not meant for public eyes. Why the whole academic system is trying to hide the misconduct?
    I would like to see policies point in place that ensures all such investigation would be made public buy the institution conducting them.
    Truth will (almost) always prevail!

  14. Glad to see the results of the investigations. The USC investigation is extremely detailed. I am not sure if that would make a difference but from the Skype messages in the FAC it appears that Li was leading the fabrications and others were following his orders.
    PS. I am not a layer but as far as I know lawsuits in California must conclude within five years of filing, and this case has been pending since June of 2018 and still does not have a date for the jury trial. How can this be?

  15. It took more than 5 years and more than $500,000 in legal fees for Sadeghi to prove his claims and expose this orchestrated scientific misconduct.
    What can be done to encourage other whistleblowers that don’t have this level of perseverance and resources to speak out?

  16. How did MBZUAI hiring committee miss Li’s history of scientific misconduct? Could they possibly still decide to hire him knowing about all this mess?
    Anyone know how many ACM SIGGRAPH papers have ever been retracted? Is there a list somewhere?

  17. I remember when Sadeghi came out with the scientific misconduct allegations and Li categorically denied everything. Back then, Sadeghi quoted Buddha that there things will not stay long hidden, the sun, the moon, and the truth.
    At that point it was actually hard to decide who is telling the truth but thanks to USC and ACM investigations all of Sadeghi’s claims have been substantiated and the truth has finally come out.
    What is the argument for not revoking the PhD degrees?

  18. I can’t think if any reason not to revoke all the phd degrees which contained these fraudulent papers in their dissertations.

  19. I would like to talk to MBZUAI’s hiring manager and se how this amount of scientific misconduct didn’t raise a red flag lol

  20. I see the note on the website that “Please only comment if not already discussed” so I had to skim through 70 comments before I felt comfortable posting mine here!!!

    I am wondering how the two appeals posted on the website came to be given that the litigation is still ongoing? What are the causes of actions that are still operating?

    Also, it would be really great for RW to do a follow-up post after the jury trial is concluded with a summary of everything that was mentioned during the public trial. Thank you.

  21. I wanted to leave a comment but I couldn’t find the link again on the main lawsuit page so I found it from the LinkedIn post.

    And I think the topic is not already discussed as requested but apologies if it is a duplicate.

    Has Retraction Watch contacted UC Berkeley to confirm that Li was not hired at the university? And that the Distinguished Fellow title is indeed not from the university. In my opinion if that is the case then it would be an undeniable misrepresentation on Hao Li’s resume as mentioned in the article.

    Thank you RW and Ellie.

  22. I don’t think the following topic had been discussed and would like to thank everyone for the all thoughtful comments;
    If ACM has any federal funding sources then all is investigating muddy be made public according to the government sunshine act.
    I personally would like to see the full investigation report and all findings be published by ACM itself.

  23. Corrected some typos:

    I don’t think the following topic has been discussed and would like to thank everyone for all the thoughtful comments;
    If ACM has any federal funding sources, then all its investigation reports must be made public according to the government Sunshine Act.
    I personally would like to see the full investigation report and all findings be published by ACM itself.
    Thank you.

  24. Given the prompt “⚠️ Please only comment if not already discussed ⚠️” on the website I am going to make my long comment much shorter and just say that I am truly impressed by how much work has been involved in bringing Pinscreen’s and Hao Li’s scientific fraud and misconduct into light.

    Sadeghi’s persistence and the amount of time, energy, money and mental capacity he has put in during this journey is admirable. I cannot thank enough all whistleblowers for their courageous initiatives.

  25. Iman Sadeghi is a brilliant engineer and a tenacious person with making sure Hao Li doesn’t continue his lying and “shitty cheating”. It’s interesting how some folks live their lives and trample on other people’s dreams and aspirations.

    It is not a pleasant process and it costs a lot of energy, time, money – that Iman could have made if he didn’t leave Google for this scam company. I am happy to see that finally truth prevails 🙂

  26. Thank you Dr. Sadeghi for your persistence and commitment to scientific integrity. It has been all worth it.

    Now that the appellate court has confirmed Dr. Li’s fraud and battery, it is time for MBZUAI to follow ACM and take corrective actions.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.