search

CyberAgentAILab / layout-dm

LayoutDM: Discrete Diffusion Model for Controllable Layout Generation [Inoue+, CVPR2023]
https://cyberagentailab.github.io/layout-dm
Apache License 2.0
211 stars 23 forks source link

Max Element Number #41

Closed Melon-Xu closed 4 months ago

Melon-Xu commented 4 months ago

Very excellent work! I am wondering if the memory occupied is related to the maximum number of elements in the layout.

I apply the model to another custom dataset, and the number of elements in one layout is much larger than 25. The memory is out.

Do I need to reduce the number of elements? Thank you so much!

naoto0804 commented 4 months ago

Hi, thank you for your interest in our project! You are right; the bottleneck is the maximum number of elements. The memory consumption grows quadratically as the number increases.

You have several workarounds. (i) reduce the maximum number of elements (ii) reduce batch size (note: you might search for hyper-parameters (e.g., learning rate and number of epochs)) (iii) reduce hidden dimension size in Transformer blocks

Melon-Xu commented 4 months ago

Thank you so much! I will try these workarounds.