Aligning the dataset using GPU does not use GPU, rather just eats up GPU

[taureanamir]

@davidsandberg,

The alignment code only eats up my GPU memory but doesnot use the GPU at all. What is the reason behind it? Is the reason same as https://github.com/davidsandberg/facenet/issues/461.

Currently I've set to use 50% of my GPU.

[MaartenBloemen]

This is normal Tensorflow behavior.

By default, TensorFlow maps nearly all of the GPU memory of all GPUs (subject to CUDA_VISIBLE_DEVICES) visible to the process.

The low GPU utilisation is as stated in the issue you referenced, because most of the code that does the aligning happens on the CPU rather than the GPU.

[Richard-wang85]

Hello! Have you solved this problem successfully? Could you tell me how to solve this problem? I will appreciate it if you could give me some advise!