Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

training speed #70

Open
crazyzyz opened this issue Oct 28, 2024 · 1 comment
Open

training speed #70

crazyzyz opened this issue Oct 28, 2024 · 1 comment

Comments

@crazyzyz
Copy link

I used 25 photos for training and set the height and width of the input image to 1080×1920 in the colmap_eval.yaml file. The number of training iterations is 40000 using RTX3080. The study found that the training time takes 50 minutes. Is this a normal training time? Looking forward to your reply, which is very important to me
1730085479016

@TQTQliu
Copy link
Owner

TQTQliu commented Oct 28, 2024

Hi, on a single RTX3090, when the image resolution is 512x640 and the batch_size is 1, it takes about 5 minutes per epoch(1000 iteration). Therefore, when the number of training iterations is 40000, it takes about 200 minutes. However, your image resolution is 1080×1920, and the training time is expected to exceed 200 minutes. But from your description and uploaded image, it only takes 50 minutes, which is strange. Besides, our model requires that H and W be integer multiples of 32. Therefore, if the height in the configuration file is set to 1080, an error will be reported. To sum up, I cannot judge whether 50 minutes is a normal time for your case. If you can provide more information, such as where specifically you modified the configuration, I would be willing to help you further.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants