vision_blender
vision_blender copied to clipboard
Possible better data storage?
I was messing around with making the file size of my dataset as small as possible, and realized that signed integers (-128 to 127) are ideal for storing the optical flow data. I don't know exactly how your saving scheme works but in my own testing, an NPY file containing a single channel was around 4 MB and after converting to signed ints it ended up being 500 KB.
Of course this may not be useful to you at all but wanted to share!
OK I actually may have been mistaken here. Think i had a wrong assumption
That's a great idea! I need to further investigate it.
I am just wondering what would happen if an object moved more that 127 pixels from one frame to the next frame. But I could still use a signed int, just with a max value higher than the resolution of the image to be rendered. Do you agree with this?