Add operator YOLO dataset quality audit

2026-07-09 01:45:52 +02:00
parent 65808ac994
commit 5898e548c2
7 changed files with 620 additions and 0 deletions
@@ -0,0 +1,160 @@
+from __future__ import annotations
+
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+def test_operator_yolo_dataset_quality_audit_reports_dataset_risks(tmp_path: Path) -> None:
+    script_path = ROOT / "scripts" / "audit_operator_yolo_dataset_quality.py"
+    assert script_path.exists()
+
+    dataset_dir = tmp_path / "yolo-dataset"
+    labels_train = dataset_dir / "labels" / "train"
+    labels_val = dataset_dir / "labels" / "val"
+    labels_train.mkdir(parents=True)
+    labels_val.mkdir(parents=True)
+
+    (labels_train / "geel_000.txt").write_text(
+        "0 0.500000 0.500000 0.010000 0.010000\n"
+        "0 0.250000 0.250000 0.100000 0.100000\n",
+        encoding="utf-8",
+    )
+    (labels_train / "postel_bos_000.txt").write_text("", encoding="utf-8")
+    (labels_train / "postel_bos_000_hn01.txt").write_text("", encoding="utf-8")
+    (labels_val / "turnhout_000.txt").write_text(
+        "0 0.600000 0.600000 0.080000 0.080000\n",
+        encoding="utf-8",
+    )
+
+    summary_path = dataset_dir / "yolo_tile_dataset_summary.json"
+    summary_path.write_text(
+        json.dumps(
+            {
+                "status": "ready",
+                "dataset_yaml": str(dataset_dir / "dataset.yaml"),
+                "output_dir": str(dataset_dir),
+                "class_names": ["building"],
+                "tile_size": 160,
+                "stride": 80,
+                "negative_keep_ratio": 1.0,
+                "background_negative_repeat": 2,
+                "min_label_px": 2,
+                "source_sample_count": 3,
+                "tile_count": 4,
+                "positive_tile_count": 2,
+                "negative_tile_count": 2,
+                "skipped_negative_tile_count": 0,
+                "label_count": 3,
+                "train_tile_count": 3,
+                "val_tile_count": 1,
+                "tiles": [
+                    {
+                        "sample_slug": "geel",
+                        "sample_role": "reference",
+                        "split": "train",
+                        "tile_index": 0,
+                        "repeat_index": 0,
+                        "kept": True,
+                        "label_path": str(labels_train / "geel_000.txt"),
+                        "label_count": 2,
+                        "is_negative": False,
+                        "is_repeated_background_negative": False,
+                    },
+                    {
+                        "sample_slug": "postel_bos",
+                        "sample_role": "background_candidate",
+                        "split": "train",
+                        "tile_index": 1,
+                        "repeat_index": 0,
+                        "kept": True,
+                        "label_path": str(labels_train / "postel_bos_000.txt"),
+                        "label_count": 0,
+                        "is_negative": True,
+                        "is_repeated_background_negative": False,
+                    },
+                    {
+                        "sample_slug": "postel_bos",
+                        "sample_role": "background_candidate",
+                        "split": "train",
+                        "tile_index": 1,
+                        "repeat_index": 1,
+                        "kept": True,
+                        "label_path": str(labels_train / "postel_bos_000_hn01.txt"),
+                        "label_count": 0,
+                        "is_negative": True,
+                        "is_repeated_background_negative": True,
+                    },
+                    {
+                        "sample_slug": "turnhout",
+                        "sample_role": "reference",
+                        "split": "val",
+                        "tile_index": 2,
+                        "repeat_index": 0,
+                        "kept": True,
+                        "label_path": str(labels_val / "turnhout_000.txt"),
+                        "label_count": 1,
+                        "is_negative": False,
+                        "is_repeated_background_negative": False,
+                    },
+                ],
+            }
+        ),
+        encoding="utf-8",
+    )
+
+    output_dir = tmp_path / "audit"
+    result = subprocess.run(
+        [
+            sys.executable,
+            str(script_path),
+            "--summary-path",
+            str(summary_path),
+            "--output-dir",
+            str(output_dir),
+            "--min-positive-samples",
+            "3",
+            "--min-val-positive-samples",
+            "2",
+            "--max-repeated-negative-share",
+            "0.25",
+            "--min-median-box-area",
+            "0.02",
+            "--max-small-box-share",
+            "0.25",
+        ],
+        cwd=ROOT,
+        check=True,
+        capture_output=True,
+        text=True,
+    )
+
+    assert "Operator YOLO dataset quality audit passed" in result.stdout
+
+    report = json.loads(
+        (output_dir / "operator_yolo_dataset_quality_audit.json").read_text(encoding="utf-8")
+    )
+    assert report["status"] == "needs_attention"
+    assert report["sample_count"] == 3
+    assert report["positive_sample_count"] == 2
+    assert report["background_sample_count"] == 1
+    assert report["train_negative_tile_count"] == 2
+    assert report["repeated_background_negative_tile_count"] == 1
+    assert report["label_stats"]["parsed_label_count"] == 3
+    assert report["label_stats"]["invalid_label_count"] == 0
+
+    warning_codes = {warning["code"] for warning in report["warnings"]}
+    assert "positive_sample_count_below_gate" in warning_codes
+    assert "val_positive_sample_count_below_gate" in warning_codes
+    assert "repeated_background_negative_share_above_gate" in warning_codes
+    assert "median_box_area_below_gate" in warning_codes
+    assert "small_box_share_above_gate" in warning_codes
+
+    markdown = (output_dir / "operator_yolo_dataset_quality_audit.md").read_text(encoding="utf-8")
+    assert "Operator YOLO Dataset Quality Audit" in markdown
+    assert "Label Quality" in markdown
+    assert "positive_sample_count_below_gate" in markdown