Here's a short description of all the desired curriculums for @blakeharrison-ai to generate. If you'd like a YAML specification from me for this curriculum I can generate that just let me know. Otherwise I'll leave the curriculum definitions to @blakeharrison-ai. If you ensure curriculum generation follows the description in #1076 it would make ADAM's processing much easier.
I've titled each section of the curriculum descriptions with the desired turn-around time for the curriculum. If any of these timelines are too aggressive please let me know.
By April 1st, 2022
[x] 15 samples each of all 20 objects in the curriculum (10 for training and 5 for testing)
[x] 15 samples each of currently implemented actions (10 for training and 5 for testing)
By April 13th, 2022
[ ] High Generation contrastive learning curriculum for objects (100 samples per object)
By April 29th, 2022
[x] 10 samples each of all 20 actions in the curriculum
By May 11th, 2022
[ ] High Generation contrastive learning curriculum for actions (100 samples per action)
June (?) (After month 9 deliverable?)
[ ] A series of scenes from a complex scene which covers the majority of the 20 objects and actions in our set (M12 stretch testing goal / demonstration decode)
The goal with any training curriculum request is to generate a large set that we can always take a subset to work on for specific debugging for faster experimentation.
Here's a short description of all the desired curriculums for @blakeharrison-ai to generate. If you'd like a YAML specification from me for this curriculum I can generate that just let me know. Otherwise I'll leave the curriculum definitions to @blakeharrison-ai. If you ensure curriculum generation follows the description in #1076 it would make ADAM's processing much easier.
I've titled each section of the curriculum descriptions with the desired turn-around time for the curriculum. If any of these timelines are too aggressive please let me know.
By April 1st, 2022
By April 13th, 2022
By April 29th, 2022
By May 11th, 2022
June (?) (After month 9 deliverable?)
The goal with any training curriculum request is to generate a large set that we can always take a subset to work on for specific debugging for faster experimentation.