Question about 3DGS training

[Suho0922]

I have some question of details in training 3DGS. could you please answer the questions?

1. do you have plan to release the 3DGS training version?

2. do you know how many images or which images are used for LORA that you use for training 3DGS?
   (more specifically it uses single image or multi-view image that contains front, back, side view of the person)

thank you for reading ☺️

[2y7c3]

1. We released the 3DGS training codes, thank you for your attention.
2. We use pre-trained LoRA to load the model (you can find various LoRA weights on Civitai). During the training process, LoRA is primarily used for training the adversarial network, specifically for the discriminator's recognition of negative samples.

[Suho0922]

thank you for answering question!! and thank your for releasing 3DGS version! I checked Civitai site ti find Lora weights, however a majority of lora weights are based on Stable diffuision v1.5. (also I can't fined Emma Stone, Taylor Swift lora for SDv2.1) however your work is based on SD v2.1 right?  is it fine to use lora weights based on SD v1.5 with SD v2.1?  Thank you!

…

-----Original Message-----From: "2y7c3" ***@***.***>To: "2y7c3/ASD" ***@***.***>; Cc: "Suho0922" ***@***.***>; "Author" ***@***.***>; Sent: 2024-09-27 (금) 13:22:50 (UTC+09:00)Subject: Re: [2y7c3/ASD] Question about 3DGS training (Issue #7) We released the 3DGS training codes, thank you for your attention. We use pre-trained LoRA to load the model (you can find various LoRA weights on Civitai). During the training process, LoRA is primarily used for training the adversarial network, specifically for the discriminator's recognition of negative samples. —Reply to this email directly, view it on GitHub #7 (comment), or unsubscribe https://github.com/notifications/unsubscribe-auth/ALINB5ETMOLDVZQZ63QJLQTZYTMRPAVCNFSM6AAAAABO4NSN42VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNZYGM3TEOJXGM.You are receiving this because you authored the thread.Message ID: ***@***.***>