minImages Training Variable about df HOT 6 CLOSED

When testing I had set minImages = 12000#min(len(images_A),len(images_B))*20 in my train.py file. However when trying to run training again, this is the output/error I get. Is 12000 not feasible for the memory usage that it would take up? I noticed that in my training with only 2000, when looking at the preview I don't see images near the end of the data set.

python train.py
running
loading C:/df/project1/data/A: 100%| 1635/1635 [00:06<00:00, 264.47it/s]
loading C:/df/project1/data/B: 100%| 11819/11819 [00:55<00:00, 212.27it/s]
Images A (1508, 256, 256, 4)
Images B (10332, 256, 256, 4)
Traceback (most recent call last):
  File "train.py", line 32, in <module>
    images_B = images_B/255.0
MemoryError

from df.

minImages = max(len(images_A),len(images_B))

from df.

That gives back the same MemoryError.

from df.

You probably don't have enough memory to load ~12k images.

In my local I changed that to max(len(images_A),len(images_B)) just as ytn2001 did

And then I added a sanity check that if minImages is greater than 6000 log a warning and set it to 6000

from df.

You use 6000 now? over 4000? Do you know if it's the VRAM or RAM that's the limitation. I'm guessing if you are using 6000 images, then you'd have 12GB of vram.

from df.

This is a limitation around keeping all of the images in memory (RAM).

from df.