code wiki / _hdl_build / nx_cross_genre.nx
nx_cross_genre.nx
buildroot/runtime/_hdl_build/nx_cross_genre.nx
about
nx_cross_genre.nx -- "cross genres WITH fidelity": reproduce a source (e.g. an anime frame) in a
different genre (e.g. photoreal) while PRESERVING the content -- pose, identity, composition,
action. The proof is a MEASUREMENT, content-neutral and genre-pair-agnostic (anime<->realism,
photo<->painting, 2D<->3D), built on the classic CONTENT/STYLE decomposition (Gatys 2015):
- CONTENT fidelity: structure/pose/identity preserved, STYLE-INVARIANT. Must stay HIGH.
- STYLE match: does the output read as the TARGET genre? Must reach the target.
- STYLE residual: does it still read as the SOURCE genre? Should be LOW (the genre crossed).
A cross-genre reproduction SUCCEEDS iff content is preserved AND the target genre is reached.
This is the scorecard that proves the system can carry a subject across genres at >= our current
realism level, without quietly losing the subject. license_tier: ORIGINAL Refs: Gatys et al.
neural style transfer (content loss vs style loss); domain-translation fidelity.
dependencies 1 imports · 1 importers
imports: nx_syscalls.nx
imported by: nx_cross_genre_test.nx
structs
| none |
consts
| 15 | const CG_SUCCESS: i64 = 0 // content preserved AND target genre reached |
| 16 | const CG_LOST_CONTENT: i64 = 1 // genre crossed but the subject/structure was wrecked |
| 17 | const CG_WRONG_STYLE: i64 = 2 // content kept but it did not reach the target genre |
functions
| 20 | func cg_verdict(content_fid: i64, style_target: i64, content_floor: i64, style_floor: i64) -> i64 |
| 27 | func cg_crossed_with_fidelity(content_fid: i64, style_target: i64, content_floor: i64, style_floor: i64) -> i64 |
| 34 | func cg_genre_moved(style_target: i64, style_source_residual: i64, style_floor: i64) -> i64 called by 1: main |
| 42 | func cg_quality(content_fid: i64, style_target: i64) -> i64 called by 1: main |