Running main.py looks like doing training?

[darthjaja6]

This shows up in log, I wonder if I did something wrong or just misunderstood. Isn't it supposed to be an inference?

[Phylliida]

TLDR: you're running inference for a single 3D model, it's just very slow

I'm not a dev for this project but how this works is it starts with a rough shape given by Point-E, then uses stable diffusion (by default, u can change this in base.yaml config guidance) from different perspectives to guide Gaussian Splatting into the right shape. Their way of doing it takes a lot of iters to converge for a single 3D model (it takes me ~3 hours on a RTX 3090 to do 15000 steps). You can see the progress in outputs/\<date>/\<date>/checkpoints/\<prompt>/\<date>/\<id>/eval, look in the image and video folders.

How frequently it shows you progress pictures/video is in base.yaml in eval, image_period and video_period

[darthjaja6]

@Phylliida Well explained, thank you!

[genitrix]

TLDR: you're running inference for a single 3D model, it's just very slow

I'm not a dev for this project but how this works is it starts with a rough shape given by Point-E, then uses stable diffusion (by default, u can change this in base.yaml config guidance) from different perspectives to guide Gaussian Splatting into the right shape. Their way of doing it takes a lot of iters to converge for a single 3D model (it takes me ~3 hours on a RTX 3090 to do 15000 steps). You can see the progress in outputs///checkpoints////eval, look in the image and video folders.

How frequently it shows you progress pictures/video is in base.yaml in eval, image_period and video_period

Hi @Phylliida, can you get the same result from the paper? I ran the training successfully but I got a result very different from my prompt.