One copy ten, CVPR oral is accused of plagiarizing a lot

Top meeting of artificial intelligence CVPR 2022 Last day , A key paper was unexpectedly popular , But not in a serious way .
On the day ,YouTube A new account appears on , Its purpose is very simple , Just to testify against a CVPR 2022 Oral The paper is suspected of plagiarism .

2022 Year of CVPR The conference The number of submissions reached 8161 Share , share 2064 Papers were received , The receiving rate is 25.28%, In the received papers , Yes 342 Were selected as Oral（ Oral reports ） The paper .

This article is accused of plagiarism Oral The paper is entitled 「E2V-SDE: From Asynchronous Events to Fast and Continuous Video Reconstruction via Neural Stochastic Differential Equations」, By the authors from Seoul University and Korea Institute of science and technology .

Address of thesis ：https://arxiv.org/abs/2206.07578
This paper presents a new model for event camera .

The event camera responds asynchronously and independently to the brightness change of each pixel in the scene . These cameras have distinctive features ： High dynamic range (HDR)、 High time resolution and low power consumption . The results of the event camera should be treated as an alternative representation of the computer vision task . However , Event cameras are susceptible to noise , And poor performance in areas with few events . In recent years , Many researchers try to reconstruct video from events . However , Due to the lack of time information from irregular and discontinuous data , Many jobs can not provide high-quality video .

In this paper , The author proposes a method called E2V-SDE The new model of , Its dynamics is determined by stochastic differential equations (SDE) Control... In potential space .E2V-SDE The image can be reconstructed quickly in any time step , And make actual predictions about invisible data .

Besides , The research claims to have successfully adopted a variety of image synthesis techniques to improve image clarity and time consistency . Through extensive experiments on simulated and real scene data sets , The researchers verified E2V-SDE The model is better than... Under various video reconstruction settings SOTA Method . In terms of image quality ,LPIPS Scores improved by up to 12%, Reconstruction speed ratio ET-Net fast 87%.

Over a period of 7 branch 16 Seconds of video , The whistleblower listed a dozen papers suspected of being plagiarized , And marked the specific paragraphs suspected of plagiarism .

These paragraphs are widely distributed in Introduction、Preliminaries、Method And so on .

E2V-SDE Thesis paragraph （ come from Introduction chapter ）.

Suspected of being plagiarized 《Latent ODEs for Irregularly-Sampled Time Series》 The paragraph .

E2V-SDE Thesis paragraph （ come from Preliminaries chapter ）.

Suspected of being plagiarized 《Continuous Latent Process Flows》 The paragraph .

E2V-SDE Thesis paragraph （ come from Preliminaries chapter ）.

Suspected of being plagiarized 《Continuous Latent Process Flows》 The paragraph .

E2V-SDE Thesis paragraph （ come from Method chapter ）.

Suspected of being plagiarized 《Vid-ODE: Continuous-time Video Generation with Neural Ordinary Differential Equation》 The paragraph .
Suspected plagiarized papers include but are not limited to those mentioned in the video list （ The list has duplicates ）：

The paper 「 Plagiarism 」 So obvious , Besides denouncing the author , They have also roast CVPR As the top computer vision will review the degree of rigor .
Of course , The open box video takes the ironic route , Netizens' comments were also full of a happy atmosphere ：

Picture source ： Zhihu answer https://www.zhihu.com/question/539378810/answer/2543500074
According to the reddit Netizens revealed , The first author of the paper has responded to this incident （ It may have been deleted later ）, He said that his research is different from the previous articles in many aspects , Some of the paragraphs mentioned in the video are actually quotes , He needs time to verify the facts .

But soon , The first author was by his teammates 「 sell 」 了 . In a post of apology , One of the co authors said , I'm sorry I didn't do my best to check 、 Obligation to verify , The plagiarism of the first author was not discovered in advance , at present , Their mentor is contacting CVPR The chairman withdrew the paper .

Then the first author confirmed that the article had been withdrawn .

Besides the bad influence of the article itself , The co-author's behavior of putting the blame on the first author has also attracted criticism from everyone .
「 Share only the glory , Not liable , There's such a good thing ？」

thus it can be seen , Many co authors may not have read the paper at all .

It is hoped that after verification by the conference Party , Events can be handled properly .
Reference Content ：https://www.zhihu.com/question/539378810https://www.reddit.com/r/MachineLearning/comments/vjkssf/d_how_to_copy_text_from_more_than_10_previously/