CUDA out of memory - Githubissues

mli0603 / stereo-transformer

Revisiting Stereo Depth Estimation From a Sequence-to-Sequence Perspective with Transformers. (ICCV 2021 Oral)

Apache License 2.0

659 stars 107 forks source link

CUDA out of memory #83

Open suriya030 opened 1 year ago

suriya030 commented 1 year ago

While running the pre trained model code in Colab/local device getting the error of ' CUDA out of memory '

suriya030 commented 1 year ago

I even tried to increase attention stride to 4 , at max I can infer result on google Colab 2 images without completely filling the gpu .

mli0603 commented 1 year ago

Hi @suriya030

The memory consumption depends on various factors. Likely your input image is too high resolution. I assume you are not using the KITTI images by default.

To run on high res images, you can

Increase the attention stride as you did. Higher strides may help.
Use the lightweight implementation in the branch here