-
Notifications
You must be signed in to change notification settings - Fork 0
/
TODO.txt
38 lines (18 loc) · 1.32 KB
/
TODO.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Various things that need to be done, yes I'm too lazy to use a ticket tracking system for this:
- search for #TODO or # TODO for spots in code with notes
NOTES:
- torch transforms are faster, but dont seem to work out as well as the kornia augmentations
- look at ascend, and see if we really want a sum of errors when using multiple prompts... ( changed to sum of squares )
TODOs:
- clean up how we modify the genJob from commands.. its hackinsh and terrible
- make masked prompt thing work, clean up the masked prompt stuff in GenerateJob
- look at nvidias torch-tensorRT for speed gains
- fix the horrible constructors i made to get rid of global variable dependency on argparse things
- refactor more, to clean up the wreckage of code. think about making a seperate generation job for spatial stuff,
and moving some things from the hallucinator class into generation job, just...clean everything now that its sorted better
- change how the image modifications that change the actual image being fed into the taming-transformers network get done
so that its possible to do multiple in the same frame, and only re-encode & recreate the optimizers once
- pop off mods as they are finished, to reduce iterations
- profile this thing to get more speed, must go faster
DONE:
- dont rely on argparse stuff across classes