Cuda appears to be in use even when setting gpu=False

I noticed this because I was already using up 95% of my GPU memory for another task. Then when I ran reader.readtext I got a RuntimeError: cuda runtime error (2) : out of memory.

This is how I create reader:

self.reader = easyocr.Reader(['en'], gpu=False)

and I can confirm that I get the following warning:

Using CPU. Note: This module is much faster with a GPU.

[rkcosmos]

thanks for the report, we’ll investigate this issue.

[jose-alanaai]

Just to add to this, in my case, if using the GPU, after loading the model, nvidia-smi reports 810MiB. This amount of course increases when doing inference and is dependent on the image size and orientation although I think you can bound it by setting canvas_size in the readtext() calls.

When setting gpu=False, after loading the model things seem fine but once you run inference on an image, the process takes 671MiB of VRAM. The only way I found to actually stop this from happening is to hide the GPU by setting the environment variable CUDA_VISIBLE_DEVICES to an empty string, e.g., os.environ['CUDA_VISIBLE_DEVICES'] = ''.

Hope this helps.