I am wondering if I can tweak the used memory of my graphics card (GTX 1060) by decreasing the quality of my used images.
The average size of my images is ~1 to around ~2 MB on 1920x1080 resolution. I am using Network size of 768 because the things to predict are less visible on lower network sizes.
Now I want to optimize this by painting the most of the images with the color Black to make the images less big in file size. With this I can reduce the image size by 1900% (to around 50 KB for each image). This is based on the bounding boxes in the training images which are not blacked out.
Will this reduce the used memory size of my graphics card or is the used memory not really affected by the used image size?