HLD map

[LuciusPertis]

This is the highest view

• [ ] film preprocessing
• [ ] #10
• [ ] #11
• [ ] generate video pieces
• [ ] video editor s/w specific export

[LuciusPertis]

discussed with ChatGPT, the algorithms for search required … k-NN, LSH (hashing), t-SNE + manifold learning algorithm; not much said about pre-sorting the images.

now thinking that the first-lvl sequences extracted from the video are small continuous manifolds themselves. So, the pose-space available is not point-noise like (what I was imagining), its like a forest (smooth forest ? lol). Like flow of nerves, divergent areas are like profiles of faces (front, left right), and scattered trees at odd poses.

hmmm ... Also, my goal is to have multiple scenes contributing to a single manifold, so I should want/expect a lot of overlapping in the pose-space.

[LuciusPertis]

ingest #10

this module work is to convert the video file into usable ds, face_frame.obj. This is further made into sequences which are cut and stitched by the sequencing module. you can say ingest just give the most basic sequences.

[LuciusPertis]

sequencing #11

This will have the power to look into the raw_sequence of actors in the video and slice and cut and stitch frames and micro sequences into what manifold is requested by the app. functions like pose stitching and pose searching etc etc