When using the Multi-dimensional Balanced MSE, there is a memory error

[BowenYang]

When using the Multi-dimensional Balanced MSE for pixel-wise loss，there is an error：not enough memory: you tried to allocate 9663676416 bytes. Buy new RAM!

[Jiawei Ren]

What is your feature dimension? The loss requires $d^2$ memory for $d$-dimensional features, so high dimensional feature may lead to OOM.

[BowenYang]

oh，thank you. I used a dense layer to reduce dimensions to solve the memory issue. However, I encountered a new problem: an error occurred in the line of code that calls torch's MVN, reporting a RuntimeError: CUDA error: no kernel image is available for execution on the device. I look forward to your reply.

[Jiawei Ren]

Hi, I am not sure how to solve this issue, you may look for solutions in pytorch forums.

[BowenYang]

I had solved this problem by reducing dimensions.Thank you.