[Pelayu]

In regards to research I'm struggling to imagine useful applications of being able to create such raw files. Also, as it's a general term for a file containing raw sensor data, it would be different across sensors.

In any case, perhaps some kind of hardware-signed cryptography scheme on the files from the camera could be used in lieu of this development in the future.

[tecleandor]

The most obvious application would be faking criminal exhibits. Fake photos prrofing something

RAW files generated "backwards" do already exist. For this, police and forensics have been using special cameras with hardware signatures for raw files for around 15 years, but Canon's and Nikon's system was broken by Elcomsoft like 12 years ago. Sony is trying something similar in their A7 IV, but I don't know how serious it is, and I think it's not as oriented to forensics as Canon and Nikon was.

https://www.elcomsoft.com/news/428.html

https://www.prnewswire.com/news-releases/elcomsoft-claims-to...

https://m.dpreview.com/news/5658217744/sony-a7-iv-gets-anti-...

[brucethemoose2]

Synthetic dataset creation for training data?

"Editable" artificial images?

I'll admit those are stretches. Anyway, in addition to a digital signing scheme, these competitions are gonna need more editing restrictions if they want to exclude the use of diffusion models.