Feature/gaussian point model - #22

Open

sridhs21 wants to merge 11 commits into

mainfrom

feature/gaussian-point-model

Commits on Jun 17, 2026

Add per-frame xpts CSV emission to run_hessian_and_build_cache.py (row, col, class columns for X/Omax/Omin points), with a backfill pass that writes CSVs for already-cached frames without re-runnin…
Swaroop Sridhar
committed
d5f92ad

Commits on Jun 24, 2026

Add point_metrics.py for scoring point-list predictions against ground truth at the point level. match_points takes two coordinate arrays and a matching radius, builds a grid index on the gt side, …
Swaroop Sridhar
committed
9b9290a

Commits on Jun 25, 2026

Add backfill_xpts_csv.py and predict_points.py, the two ways we get point-list CSVs onto disk. backfill_xpts_csv.py walks a cache directory and writes per-frame {N}_xpts.csv files from existing {N}…
Swaroop Sridhar
committed
251ae04

Commits on Jul 7, 2026

Add score_point_predictions.py, the aggregator that ties the point-list scoring together. Takes a predictions root directory (subdirs per dataset), loads the predicted xpts CSVs with their confiden…
Swaroop Sridhar
committed
8f4d24b

Commits on Jul 24, 2026

Bring Gaussian heatmap training code (gaussian_xpoints_mask, FocalHeatmapLoss, --gaussianSigma/--targetType/--heatmap* args) onto the point-list branch so a single working tree can both train at va…
Swaroop Sridhar
committed
6252c80

Commits on Aug 20, 2026

move peak extraction into point_metrics and vectorize the connected-component NMS
Swaroop Sridhar
committed
3e2bf69
use the shared extract_peaks in predict_points and thread the second PKPM run through cache build, prediction, and scoring
Swaroop Sridhar
committed
85fabba
add a gaussian-aware optuna tuner that maximizes point-level F1
Swaroop Sridhar
committed
706fd4d

Commits on Sep 4, 2026

derive predict/score dataset paths from RC_EXTRACT_DIR/RC_CACHE_BASE instead of hardcoding them
sridhs21
committed
deb3d38

Commits on Sep 11, 2026