Support for flame2023?

[jdeepee]

Awesome project all. Incredible to see faster than realtime mesh output!

Do you have any plans to train pear for flame2023 vs the current 2020 implementation. Would be fantastic for one of my projects which manipulates gaussians using flame2023 params.

[WuJH2001]

Hi @jdeepee, thank you for your interest!
Based on our experiments, we did not observe a significant difference in expressiveness between FLAME 2023 and FLAME 2020; the main differences are in some parameter settings.
Therefore, we currently do not plan to train PEAR with FLAME 2023 in the short term.
That said, if you are interested in training it yourself, you are very welcome to join us — we can provide initial data support to help train PEAR with FLAME 2023.

[jdeepee]

I would be happy to train for you. Let me know how I can access the data and training code and I will get to work.

[WuJH2001]

Hi @jdeepee, sorry for the delay caused by our earlier misestimation. Due to the large data volume and some remaining procedures that need to be completed, it may take some more time before we can release all the training data and code.

[lizhaohu]

Great! Looking forward to the release of training data. I also need flame2023 in my research

[WuJH2001]

Hi, thanks for your interest! We’re organizing the training code and plan to open-source it in about a week.