Ai Dreams Forum
Member's Experiments & Projects => AI Programming => Topic started by: ranch vermin on May 08, 2015, 09:55:03 pm
-
https://www.youtube.com/watch?v=mbgW5JDvHiE&feature=youtu.be (https://www.youtube.com/watch?v=mbgW5JDvHiE&feature=youtu.be)
this is pretty steady point tracking, now i go add my key memory system.
if a tracker is lone in its next pixel, it goes into the first key memory, written.
if a tracker ever goes to the same pixel as another tracker, then i opt to read the key thats there not write to it.
then if a tracker reads, i then decide to make a group of the trackers came to same pixel.
I have a feeling now I should be able to form a segmentation map, from looking at the connections of the groups made, where trackers came together, and in fact form a plush toy that could be inflated, but the vectors and distances are unknown.
I can mip the screen and i can say these are twice as small, and then track on this mip, and decide its twice as far away, im not sure but that might help me make it 3d.
I am going to form insensitive softbody symantics, then i become more specific about the posture later. taxonomies are always on my mind, everything gets put into taxonomies, but the thing about a taxonomy, you only have what happened specificly, that the group is always incomplete, but if I put things in a taxonomy, now if i see the insensitive, i can then apply the sensitive that i know, or i can even hybrid results.
I may decide not to bother with tracking later on, because a spacial based single shot system is better, but maybe itll turn out i need temporal to make 3d, but i dont know yet.