Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions docs/known-limits.md
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,13 @@
Approaches already measured to fail.
Do not retry one without a new idea, and add a line here when you measure a new dead end.

- Geometric-mean scale convergence can move a previously passing per-frame median outside the
fixed band as visible depth support changes. After normal convergence, the automatic fitter
retains the earlier median-passing probe closest to unit geometric mean only if the final
medians fail, keeping that probe's scale and measured evidence together. Nonconvergence still
fails; explicit scales are unchanged. Median retention does not relax the independent depth
spread gate or establish acceptable geometry.

- The fresh city tracking baseline assigned person poses to street lamps and split the main
walker across tracks. Four of five retained tracks failed the source-appearance identity audit;
202 Mask R-CNN detections lacked a pose partner. These counts do not establish how many real
Expand Down
26 changes: 23 additions & 3 deletions scripts/run_clip.py
Original file line number Diff line number Diff line change
Expand Up @@ -3084,13 +3084,22 @@ def scale_fit(self):
depth is ~1/scale0, so scale0 <- scale0 * geomean(ratio) converges in two or three probes.
"""
s, probes = self.a.scale0, []
selected_probe, selection_reason = "given", "explicit_scale"
if s is None:
best_median_probe = None
s = self.placement().get("floorMatchedScale") or 1.0
for i in range(self.a.scale_probes):
rows, fig, bands = self.diag(s, i)
ratios = [r for r, _, _ in rows]
gm = math.exp(sum(math.log(r) for r in ratios) / len(ratios))
probes.append((s, gm))
if all(SCALE_RATIO_BAND[0] <= r <= SCALE_RATIO_BAND[1] for r in ratios) and (
best_median_probe is None
or abs(math.log(gm)) < abs(math.log(best_median_probe[2]))
):
# Keep the measured evidence together. Diagnostic filenames round the
# scale, so a later probe can overwrite the same figure path.
best_median_probe = (i, s, gm, rows, bands, Path(fig).read_bytes())
say(
f" probe {i}: scale0 {s:.4f} -> ratios {[round(r, 3) for r in ratios]} "
f"(geomean {gm:.4f})"
Expand All @@ -3104,6 +3113,15 @@ def scale_fit(self):
f"{[(round(x, 4), round(g, 4)) for x, g in probes]}; pass "
f"--scale0 by hand or raise --scale-probes"
)
selected_probe, selection_reason = i, "converged_final"
if best_median_probe is not None and not all(
SCALE_RATIO_BAND[0] <= r <= SCALE_RATIO_BAND[1] for r in ratios
):
selected_probe, s, gm, rows, bands, figure_bytes = best_median_probe
ratios = [r for r, _, _ in rows]
fig = self.ctx / f"scale-fit-selected-probe-{selected_probe}.png"
fig.write_bytes(figure_bytes)
selection_reason = "earlier_median_pass_closest_log_geomean"
else:
rows, fig, bands = self.diag(s, "given")
ratios = [r for r, _, _ in rows]
Expand All @@ -3119,9 +3137,9 @@ def scale_fit(self):
if spread:
median_problem = (
f"depth ratios {[round(r, 3) for r in ratios]} are not all inside "
f"{SCALE_RATIO_BAND[0]}-{SCALE_RATIO_BAND[1]} at any single scale (geomean fit {s:.4f} "
f"gives {gm:.4f}). That is Marble geometry error or SfM drift, not a scale error -- it "
f"means the avatar reads ~{abs(1 - min(ratios)) * 100:.0f}% wrong at one end of the "
f"{SCALE_RATIO_BAND[0]}-{SCALE_RATIO_BAND[1]} at the evaluated scale {s:.4f} "
f"(geomean {gm:.4f}). This measurement alone does not rule out another scale. "
f"The depth mismatch is ~{abs(1 - min(ratios)) * 100:.0f}% at one end of the "
f"clip. Read {fig}, then either re-trim the clip to "
f"the window that fits or pass --allow-scale-spread to ship the geometric-mean fit "
f"with that caveat (gym, atrium and selfie did)."
Expand All @@ -3142,6 +3160,8 @@ def scale_fit(self):
p10=p10,
p90=p90,
probes=probes,
selectedProbe=selected_probe,
selectionReason=selection_reason,
gateOk=not spread and not wide,
medianOk=not spread,
spreadOk=not wide,
Expand Down
129 changes: 129 additions & 0 deletions scripts/tests/test_scale_fit_evidence.py
Original file line number Diff line number Diff line change
Expand Up @@ -213,3 +213,132 @@ def test_band_summary_keeps_gate_precision_and_flags_collapsed_quartiles():

def test_malformed_band_lines_are_evidence_loss_not_a_crash():
assert run_clip.distance_bands('depth-ratio-bands {"frame": 0, "bands": [', 1) is None


def probe_sequence(pipeline, monkeypatch, rows_by_probe):
"""Different measured evidence per call; reuse a filename to exercise snapshot ownership."""
calls, evidence = [], []
figure = pipeline.ctx / "rounded-scale-depth-ratio.png"

def diag(scale, probe):
index = len(calls)
rows = rows_by_probe[index]
bands = [
{"frame": frame, "bands": [{"band": 0, "p10": lo, "median": ratio, "p90": hi}]}
for frame, (ratio, lo, hi) in enumerate(rows)
]
pixels = f"measured probe {index}".encode()
figure.write_bytes(pixels)
calls.append((scale, probe))
evidence.append((rows, bands, pixels))
return rows, figure, bands

monkeypatch.setattr(pipeline, "diag", diag)
return calls, evidence, figure


@pytest.mark.parametrize("unaccepted", [False, True])
def test_retains_closest_passing_probe_with_consistent_evidence_and_failed_spread(
pipeline, monkeypatch, unaccepted
):
p = pipeline
p.a.scale_spread_unaccepted = unaccepted
rows = [
[(1.4, 1.2, 1.6)],
[(0.98, 0.94, 1.04), (0.98, 0.93, 1.05)],
[(1.01, 0.81, 1.21), (1.01, 0.85, 1.16)],
[(0.94, 0.92, 1.06), (1.06, 0.95, 1.08)],
]
calls, evidence, overwritten = probe_sequence(p, monkeypatch, rows)
assert p.go() == (0 if unaccepted else 1)
assert len(calls) == 4 # No additional diagnostic or retry.
report = json.loads((p.ctx / "scale-fit.json").read_text())
assert report == p.state.data["stages"]["_scale"]
assert report["selectedProbe"] == 2
assert report["selectionReason"] == "earlier_median_pass_closest_log_geomean"
assert report["scale0"] == pytest.approx(calls[2][0])
assert report["scale0"] == pytest.approx(0.8 * 1.4 * 0.98)
assert report["geomean"] == pytest.approx(1.01)
assert report["ratios"] == [1.01, 1.01]
assert report["p10"] == [0.81, 0.85]
assert report["p90"] == [1.21, 1.16]
assert report["distanceBands"] == evidence[2][1]
assert Path(report["figure"]).read_bytes() == evidence[2][2]
assert overwritten.read_bytes() == evidence[3][2]
assert report["medianOk"] is True
assert report["spreadOk"] is False
assert report["gateOk"] is False
assert "exceed the required spread" in report["gateError"]
assert report["figure"] in report["gateError"]
assert len(report["probes"]) == 4
assert report["probes"][3][0] == pytest.approx(calls[3][0])


def test_median_passing_final_keeps_final_even_if_earlier_spread_was_better(pipeline, monkeypatch):
rows = [[(1.01, 0.97, 1.04)], [(1.001, 0.8, 1.2)]]
calls, evidence, figure = probe_sequence(pipeline, monkeypatch, rows)
assert pipeline.go() == 1
report = json.loads((pipeline.ctx / "scale-fit.json").read_text())
assert len(calls) == 2
assert report["selectedProbe"] == 1
assert report["selectionReason"] == "converged_final"
assert report["scale0"] == pytest.approx(calls[1][0])
assert report["ratios"] == [1.001]
assert report["distanceBands"] == evidence[1][1]
assert report["figure"] == str(figure)
assert report["medianOk"] is True and report["spreadOk"] is False
assert not list(pipeline.ctx.glob("scale-fit-selected-probe-*.png"))


def test_no_acceptable_probe_keeps_final_failure_and_limits_error_claim(pipeline, monkeypatch):
calls, _, figure = probe_sequence(
pipeline, monkeypatch, [[(1.2, 1.0, 1.3)], [(0.94, 0.92, 1.08), (1.06, 0.92, 1.08)]]
)
assert pipeline.go() == 1
report = json.loads((pipeline.ctx / "scale-fit.json").read_text())
assert len(calls) == 2
assert report["selectedProbe"] == 1
assert report["scale0"] == pytest.approx(calls[1][0])
assert report["figure"] == str(figure)
assert report["medianOk"] is False and report["spreadOk"] is True
assert "at the evaluated scale" in report["gateError"]
assert "at any single scale" not in report["gateError"]


def test_acceptable_probe_does_not_bypass_nonconvergence(pipeline, monkeypatch):
pipeline.a.scale_probes = 2
calls, _, _ = probe_sequence(pipeline, monkeypatch, [[(1.02, 0.97, 1.04)]] * 2)
assert pipeline.go() == 1
assert len(calls) == 2
stages = pipeline.state.data["stages"]
assert "did not converge" in stages["scale_fit"]["error"]
assert "_scale" not in stages
assert not (pipeline.ctx / "scale-fit.json").exists()
assert not list(pipeline.ctx.glob("scale-fit-selected-probe-*.png"))


@pytest.mark.parametrize("ratio,accepted", [(1.02, True), (1.07, False)])
def test_explicit_scale_keeps_single_given_measurement(pipeline, monkeypatch, ratio, accepted):
pipeline.a.scale0 = 0.726089
calls, evidence, figure = probe_sequence(pipeline, monkeypatch, [[(ratio, 0.96, 1.08)]])
assert pipeline.go() == (0 if accepted else 1)
assert calls == [(0.726089, "given")]
report = json.loads((pipeline.ctx / "scale-fit.json").read_text())
assert report["scale0"] == 0.726089
assert report["selectedProbe"] == "given" and report["selectionReason"] == "explicit_scale"
assert report["probes"] == []
assert report["figure"] == str(figure)
assert report["distanceBands"] == evidence[0][1]
assert report["gateOk"] is accepted


def test_retained_probe_passes_gate_only_when_its_spread_also_passes(pipeline, monkeypatch):
calls, _, _ = probe_sequence(
pipeline, monkeypatch, [[(1.02, 0.96, 1.08)], [(0.94, 0.7, 1.3), (1.06, 0.7, 1.3)]]
)
assert pipeline.go() == 0
report = json.loads((pipeline.ctx / "scale-fit.json").read_text())
assert len(calls) == 2
assert report["selectedProbe"] == 0
assert report["scale0"] == 0.8
assert report["gateOk"] is True and report["gateError"] is None
Loading