What happened. Your Q1–Q13 answers of 29 Aug retuned the hedge gate. v5 adds two conjuncts (h_top ≤ 3.3 m, relief ≤ 2.2 m) and drops the area floor from 250 m2 to 100 m2, keeping stems < 1.6 / 10 m. The gate now fires on 61 parents corpus-wide (v4: 21).
Score on the truth set. v5 recall 0.904, FP 4/30, weighted 0.637 — against v4 recall 0.916 / FP 9/30 and today's detector (v0) recall 0.482 / FP 1/30. v5 gives up 1.2 points of recall and removes 5 of v4's 9 false positives.
Corpus A1B0 accepted tree points. production 14,004,123 → v4 25,572,719 → v5 22,689,838. v5 still nearly doubles today's tree paint; the 8,685,715 extra points are the recall gain, the 2,882,881 difference from v4 is what the retuned gate refuses.
What I need from you. 25 one-word calls: is each of these a hedge (must stay unpainted) or trees (must be painted)? Group A is what v5 newly refuses — if any of them is trees, the gate is too tight. Group B is what v5 keeps by a hair — if any of them is a hedge, the gate is too loose.
A cluster the frozen RF refused is normally cut into 8 m tiles and each tile re-scored (tile-and-score, rescue-only). The parent hedge gate stops that rescue when the parent reads as a clipped wall. All four conjuncts must hold:
h_top <= 3.30 m crown top above the ground (robust)
relief <= 2.20 m p90 - p10 of the crown-top surface
stems < 1.60 /10 m strong vertical stems along the band
area >= 100 m2 footprint, so slivers are not gatedv4 used area ≥ 250 m2 and had no h_top / relief conjunct; it relied on
continuity and a flat-top test instead. The change is a direct read of your 29 Aug answers:
everything you called a hedge sat under 3.3 m with a flat top, everything you called trees either broke
3.3 m or had > 2.2 m of relief in the crown surface.
These parents were partly painted as tree by v4 and are entirely unpainted by v5. BEFORE = v4, AFTER = v5; two spots per parent, placed at ±30 % of its length along its long axis. Green paint = claimed as tree; teal = medium_vegetation. Click any image for full resolution; ← / → (or swipe) steps through that card's images only, Esc closes. Each caption also links the top-down view.
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q1 — 018 c0: hedge or trees?
answer: Q1: hedge / Q1: trees / Q1: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q2 — 068 c1: hedge or trees?
answer: Q2: hedge / Q2: trees / Q2: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q3 — 121 c5: hedge or trees?
answer: Q3: hedge / Q3: trees / Q3: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q4 — 021 c0: hedge or trees?
answer: Q4: hedge / Q4: trees / Q4: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q5 — 077 c4: hedge or trees?
answer: Q5: hedge / Q5: trees / Q5: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q6 — 083 c3: hedge or trees?
answer: Q6: hedge / Q6: trees / Q6: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q7 — 061 c11: hedge or trees?
answer: Q7: hedge / Q7: trees / Q7: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q8 — 066 c8: hedge or trees?
answer: Q8: hedge / Q8: trees / Q8: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q9 — 122 c7: hedge or trees?
answer: Q9: hedge / Q9: trees / Q9: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q10 — 053 c0: hedge or trees?
answer: Q10: hedge / Q10: trees / Q10: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q11 — 076 c4: hedge or trees?
answer: Q11: hedge / Q11: trees / Q11: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q12 — 033 c0: hedge or trees?
answer: Q12: hedge / Q12: trees / Q12: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q13 — 076 c6: hedge or trees?
answer: Q13: hedge / Q13: trees / Q13: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q14 — 094 c1: hedge or trees?
answer: Q14: hedge / Q14: trees / Q14: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q15 — 108 c34: hedge or trees?
answer: Q15: hedge / Q15: trees / Q15: mixed - <where>
v5 gates this parent as a clipped band: the whole cluster stays unpainted (reject_rf_hedge_band). Under v4 it was tiled and part of it was painted as tree.
Q16 — 117 c10: hedge or trees?
answer: Q16: hedge / Q16: trees / Q16: mixed - <where>
These sit just above the gate ceiling (h_top 3.3–3.8 m), so v5 still rescues them. If one of these is a hedge to you, the ceiling has to move up.
v5 keeps this parent in the rescue path (h_top above the 3.3 m gate ceiling), so it is tiled and the accepted tiles are painted as tree - identical to v4. These are the closest survivors to the new rule.
Q17 — 094 c0: hedge or trees?
answer: Q17: hedge / Q17: trees / Q17: mixed - <where>
v5 keeps this parent in the rescue path (h_top above the 3.3 m gate ceiling), so it is tiled and the accepted tiles are painted as tree - identical to v4. These are the closest survivors to the new rule.
Q18 — 023 c7: hedge or trees?
answer: Q18: hedge / Q18: trees / Q18: mixed - <where>
v5 keeps this parent in the rescue path (h_top above the 3.3 m gate ceiling), so it is tiled and the accepted tiles are painted as tree - identical to v4. These are the closest survivors to the new rule.
Q19 — 082 c1: hedge or trees?
answer: Q19: hedge / Q19: trees / Q19: mixed - <where>
v5 keeps this parent in the rescue path (h_top above the 3.3 m gate ceiling), so it is tiled and the accepted tiles are painted as tree - identical to v4. These are the closest survivors to the new rule.
Q20 — 031 c1: hedge or trees?
answer: Q20: hedge / Q20: trees / Q20: mixed - <where>
v5 keeps this parent in the rescue path (h_top above the 3.3 m gate ceiling), so it is tiled and the accepted tiles are painted as tree - identical to v4. These are the closest survivors to the new rule.
Q21 — 096 c6: hedge or trees?
answer: Q21: hedge / Q21: trees / Q21: mixed - <where>
v5 keeps this parent in the rescue path (h_top above the 3.3 m gate ceiling), so it is tiled and the accepted tiles are painted as tree - identical to v4. These are the closest survivors to the new rule.
Q22 — 078 c9: hedge or trees?
answer: Q22: hedge / Q22: trees / Q22: mixed - <where>
v5 keeps this parent in the rescue path (h_top above the 3.3 m gate ceiling), so it is tiled and the accepted tiles are painted as tree - identical to v4. These are the closest survivors to the new rule.
Q23 — 073 c17: hedge or trees?
answer: Q23: hedge / Q23: trees / Q23: mixed - <where>
v5 keeps this parent in the rescue path (h_top above the 3.3 m gate ceiling), so it is tiled and the accepted tiles are painted as tree - identical to v4. These are the closest survivors to the new rule.
Q24 — 048 c19: hedge or trees?
answer: Q24: hedge / Q24: trees / Q24: mixed - <where>
You answered Q5 with "trees, but I see 4 trunks in the back of the spot that stay medium_vegetation". Here are the three original spots rendered on the v5 cloud, plus every tile the rescue scored.
Unchanged between v4 and v5 (h_top 2.35 m but 1.92 stems/10 m keeps it out of the gate). You reported four trunks at the back of lowrej_066_c15_s1 that stay medium_vegetation.
All 17 tiles of parent c15 in v5_corpus/segment_066/tree_clusters.csv (rows with parent_cluster_id=15). 9 accepted / 8 rejected, identical in v4 and v5. There is no RF-probability column in tree_clusters.csv (the schema stores outcome/reason only), so the reject reason is the finest grain available. Distances are metres from the tile centroid to each rendered spot; the tiles nearest _s1 are 6 (accepted, 36,857 pts), 4 (rejected, 28,351 pts), 5 and 7 (accepted), 8 (rejected, 35,809 pts). Every rejected tile has h_max_robust below 2.0 m except the two 86/95-point slivers (3.78 m / 4.97 m) that fail tree_min_points-scale mass: those two are the likeliest home of your four trunks.
| tile | outcome | n_points | h_max_robust | ExG | crown_isotropy | d to s0 | d to s1 | d to s2 |
|---|---|---|---|---|---|---|---|---|
| 0 | reject_rf_below_threshold | 18,200 | 1.873 | 0.25 | 0.7538 | 1.0 | 20.1 | 40.3 |
| 1 | vegetation_rf | 676 | 2.692 | 0.2857 | 0.7171 | 6.2 | 21.5 | 41.2 |
| 2 | reject_rf_below_threshold | 31,935 | 1.991 | 0.2663 | 0.7406 | 5.8 | 13.5 | 33.7 |
| 3 | vegetation_rf | 3,430 | 2.97 | 0.2571 | 0.9163 | 10.1 | 14.9 | 33.7 |
| 4 | reject_rf_below_threshold | 28,351 | 1.882 | 0.2375 | 1.2921 | 14.0 | 5.3 | 25.5 |
| 5 | vegetation_rf | 4,014 | 2.552 | 0.25 | 0.7207 | 16.1 | 8.9 | 26.0 |
| 6 | vegetation_rf | 36,857 | 2.072 | 0.2083 | 1.0835 | 21.9 | 2.6 | 17.6 |
| 7 | vegetation_rf | 3,886 | 2.617 | 0.223 | 0.6301 | 23.0 | 8.6 | 19.5 |
| 8 | reject_rf_below_threshold | 35,809 | 1.998 | 0.2171 | 1.0175 | 29.9 | 10.7 | 9.6 |
| 9 | vegetation_rf | 3,198 | 2.999 | 0.2222 | 1.0208 | 30.5 | 12.7 | 12.1 |
| 10 | reject_rf_below_threshold | 86 | 3.776 | 0.1659 | 0.3243 | 29.7 | 15.8 | 19.1 |
| 11 | vegetation_rf | 34,743 | 1.989 | 0.1765 | 0.7886 | 37.7 | 18.4 | 1.8 |
| 12 | vegetation_rf | 1,735 | 3.019 | 0.1695 | 1.2139 | 37.9 | 19.6 | 8.2 |
| 13 | reject_rf_below_threshold | 11,287 | 1.977 | 0.2222 | 0.6433 | 44.8 | 25.5 | 5.3 |
| 14 | vegetation_rf | 1,174 | 3.25 | 0.2 | 0.8321 | 46.0 | 27.3 | 10.1 |
| 15 | reject_rf_below_threshold | 95 | 4.97 | 0.1667 | 0.2246 | 43.1 | 26.0 | 14.7 |
| 16 | reject_rf_below_threshold | 1,158 | 1.87 | 0.25 | 0.2522 | 50.3 | 31.0 | 10.8 |
Q25 — 066 c15: hedge or trees?
answer: Q25: hedge / Q25: trees / Q25: mixed - <where>
Copy, fill in, send back. hedge / trees / mixed - <where>.
Q1:
Q2:
Q3:
Q4:
Q5:
Q6:
Q7:
Q8:
Q9:
Q10:
Q11:
Q12:
Q13:
Q14:
Q15:
Q16:
Q17:
Q18:
Q19:
Q20:
Q21:
Q22:
Q23:
Q24:
Q25:
Q-cfg: (ship v5 gate as it stands? yes / no + change)