berkus
/
wifi-densepose
mirror of https://github.com/ruvnet/wifi-densepose.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

tune(adr-120): adaptive smoothing window 7 → 15 ticks (~1.5s)

This commit is contained in:
arsen 2026-05-18 01:23:09 +07:00
parent 3e12686ae9
commit c3f00f3abf
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
v2/crates/wifi-densepose-sensing-server/src/main.rs
View File

@ -2729,10 +2729,10 @@ fn adaptive_override(state: &AppStateInner, features: &FeatureInfo, classificati
/// ADR-120 follow-up: majority-vote smoothing buffer for the adaptive /// ADR-120 follow-up: majority-vote smoothing buffer for the adaptive
/// classifier output. At the broadcast tick rate (~10 Hz) the model emits /// classifier output. At the broadcast tick rate (~10 Hz) the model emits
/// a fresh decision every ~100 ms, and adjacent decisions can disagree /// a fresh decision every ~100 ms, and adjacent decisions can disagree
/// even when reality is stable (UI flicker). We keep the last 7 ticks /// even when reality is stable (UI flicker). 15-tick window (~1.5 sec)
/// (~700 ms) and display the mode. Snappy enough for live UX, stable /// favours readability over reaction speed — sustained activity wins,
/// enough that the user can read the label without it changing mid-read. /// brief glitches don't update the display.
const ADAPTIVE_SMOOTH_WIN: usize = 7; const ADAPTIVE_SMOOTH_WIN: usize = 15;
static ADAPTIVE_LABEL_HISTORY: OnceLock<Mutex<VecDeque<String>>> = OnceLock::new(); static ADAPTIVE_LABEL_HISTORY: OnceLock<Mutex<VecDeque<String>>> = OnceLock::new();