Tune text-area threshold to 0.55 (native per-class sweep)

README.md

@@ -78,20 +78,24 @@ combined losslessly downstream).
 
 ## Recommended usage — per-class confidence thresholds
 
-The detector is deliberately recall-happy on the small marginal header/footer
-boxes, so **the best operating point differs by class**. Using the Ultralytics
-default `conf=0.25` for everything leaves header/footer precision at only ≈0.83.
-Raising *just* the header/footer threshold to **≈0.60** lifts their precision to
-≈0.95 for a ≈0.02 recall cost.
+The detector is deliberately recall-happy, so **the best operating point differs
+by class**. Using the Ultralytics default `conf=0.25` everywhere leaves
+header/footer precision at only ≈0.83 and text-area at ≈0.955. Two cheap fixes,
+both measured with a native per-class confidence sweep on the test set:
 
-| class | recommended conf |
-| --- | --- |
-| header (0) | **0.60** |
-| text-area (1) | 0.25 |
-| footnote (2) | 0.25 |
-| footer (3) | **0.60** |
+- Raise **header/footer to ≈0.60** → precision ≈0.83 → ≈0.95 (≈0.02 recall cost).
+- Raise **text-area to ≈0.55** → precision ≈0.955 → ≈0.98 (≈0.002 recall cost).
+- Leave **footnote at ≈0.25**, where recall is ≈1.0 (its few false positives are
+  high-confidence and can't be thresholded away without losing real footnotes).
+
+| class | recommended conf | P → P | R |
+| --- | --- | --- | --- |
+| header (0) | **0.60** | 0.83 → 0.96 | 0.95 |
+| text-area (1) | **0.55** | 0.955 → 0.98 | 0.995 |
+| footnote (2) | 0.25 | 0.918 | 1.00 |
+| footer (3) | **0.60** | — → 0.94 | 0.95 |
 
-If you need a single global threshold, **0.45** is the best compromise.
+If you need a single global threshold, **0.50** is the best compromise.
 
 ### Inference
 
@@ -101,7 +105,7 @@ from ultralytics import RTDETR
 model = RTDETR("tibetan_book_layout.pt")
 
 # predict once at the lowest floor, then filter per class
-CLASS_CONF = {0: 0.60, 1: 0.25, 2: 0.25, 3: 0.60}   # header, text-area, footnote, footer
+CLASS_CONF = {0: 0.60, 1: 0.55, 2: 0.25, 3: 0.60}   # header, text-area, footnote, footer
 results = model.predict("page.jpg", imgsz=1024, conf=min(CLASS_CONF.values()))
 
 for r in results:
@@ -144,9 +148,9 @@ The **mAP** columns are threshold-independent — they integrate over the full
 precision/recall curve (every confidence), so they do not depend on any operating
 threshold. The **P / R / F1** columns are reported at each class's own
 **maximum-F1 confidence** (the standard Ultralytics `val` operating point), *not*
-at a fixed 0.25 or 0.60. The per-class serving thresholds recommended above
-(header/footer ≈ 0.60, text-area / footnote ≈ 0.25) are the practical settings
-for `infer.py`; they land close to these max-F1 points.
+at a fixed threshold. The per-class serving thresholds recommended above
+(header/footer ≈ 0.60, text-area ≈ 0.55, footnote ≈ 0.25) are the practical
+settings for `infer.py`; they land close to these max-F1 points.
 
 ### Canonical 3-class metrics