search

shad0wta9 / meshavatar

Code Repository for MeshAvatar: Learning High-quality Triangular Human Avatars from Multi-view Videos (ECCV 2024)
Other
74 stars 2 forks source link

memory requirement #3

Open yejr0229 opened 4 weeks ago

yejr0229 commented 4 weeks ago

Hi, I'd like to know how much memory will it take? Is a 3090 with 24gb enough?

shad0wta9 commented 4 weeks ago

It depends on the resolution of the images and the proportion of the human body in the image. For AvatarReX dataset (1024x750), it takes around 15gb to train. You can reduce the batch_size in the config file if the memory is not enough.

ZhenyanSun commented 2 weeks ago

It depends on the resolution of the images and the proportion of the human body in the image. For AvatarReX dataset (1024x750), it takes around 15gb to train. You can reduce the batch_size in the config file if the memory is not enough.

Thanks for sharing your great work. If we resize the img, the intrinsic matrices be scaled too. I find you put that k[:2]X0.5 in other issue, I was wondering why the k[4] keep no change, should it also be k[4]x0.5?

shad0wta9 commented 2 weeks ago

@ZhenyanSun K[:2] means the top 2 rows of the intrinsic matrix

ZhenyanSun commented 2 weeks ago

@ZhenyanSun K[:2] means the top 2 rows of the intrinsic matrix

thx very much，I thought it was k[9].