Add operator YOLO label QA contact sheets

2026-07-11 21:34:22 +02:00
parent 8cb435361d
commit fbccf8322e
7 changed files with 682 additions and 0 deletions
@@ -61,6 +61,10 @@ def test_all_in_one_dockerfile_copies_operator_scripts_for_runtime_use() -> None
    assert "COPY scripts/prepare_operator_real_data_samples.py /app/scripts/prepare_operator_real_data_samples.py" in dockerfile
    assert "COPY scripts/export_operator_yolo_tile_dataset.py /app/scripts/export_operator_yolo_tile_dataset.py" in dockerfile
    assert "COPY scripts/audit_operator_yolo_dataset_quality.py /app/scripts/audit_operator_yolo_dataset_quality.py" in dockerfile
+    assert (
+        "COPY scripts/render_operator_yolo_label_qa_contact_sheets.py "
+        "/app/scripts/render_operator_yolo_label_qa_contact_sheets.py"
+    ) in dockerfile
    assert "COPY scripts/train_operator_yolo_detector.sh /app/scripts/train_operator_yolo_detector.sh" in dockerfile


@@ -0,0 +1,162 @@
+from __future__ import annotations
+
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+from PIL import Image
+
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+def test_operator_yolo_label_qa_contact_sheets_render_visual_artifacts(tmp_path: Path) -> None:
+    script_path = ROOT / "scripts" / "render_operator_yolo_label_qa_contact_sheets.py"
+    assert script_path.exists()
+
+    dataset_dir = tmp_path / "yolo-dataset"
+    image_train = dataset_dir / "images" / "train"
+    image_val = dataset_dir / "images" / "val"
+    labels_train = dataset_dir / "labels" / "train"
+    labels_val = dataset_dir / "labels" / "val"
+    image_train.mkdir(parents=True)
+    image_val.mkdir(parents=True)
+    labels_train.mkdir(parents=True)
+    labels_val.mkdir(parents=True)
+
+    for path, color in (
+        (image_train / "dense_000.png", (120, 130, 140)),
+        (image_train / "invalid_000.png", (80, 100, 120)),
+        (image_val / "missing_000.png", (90, 120, 90)),
+        (image_val / "negative_000.png", (50, 50, 55)),
+    ):
+        Image.new("RGB", (64, 64), color=color).save(path)
+
+    (labels_train / "dense_000.txt").write_text(
+        "0 0.500000 0.500000 0.500000 0.500000\n"
+        "0 0.250000 0.250000 0.250000 0.250000\n",
+        encoding="utf-8",
+    )
+    (labels_train / "invalid_000.txt").write_text(
+        "0 0.500000 0.500000 0.300000 0.300000\n"
+        "not-a-valid-yolo-row\n",
+        encoding="utf-8",
+    )
+    (labels_val / "negative_000.txt").write_text("", encoding="utf-8")
+
+    missing_label_path = labels_val / "missing_000.txt"
+    summary_path = dataset_dir / "yolo_tile_dataset_summary.json"
+    summary_path.write_text(
+        json.dumps(
+            {
+                "status": "ok",
+                "dataset_yaml": str(dataset_dir / "dataset.yaml"),
+                "output_dir": str(dataset_dir),
+                "class_names": ["building"],
+                "tile_size": 64,
+                "stride": 64,
+                "tiles": [
+                    {
+                        "sample_slug": "dense",
+                        "sample_role": "reference",
+                        "background_category": "reference_aoi",
+                        "split": "train",
+                        "tile_index": 0,
+                        "kept": True,
+                        "image_path": str(image_train / "dense_000.png"),
+                        "label_path": str(labels_train / "dense_000.txt"),
+                        "label_count": 2,
+                        "is_negative": False,
+                    },
+                    {
+                        "sample_slug": "invalid",
+                        "sample_role": "reference",
+                        "background_category": "reference_aoi",
+                        "split": "train",
+                        "tile_index": 1,
+                        "kept": True,
+                        "image_path": str(image_train / "invalid_000.png"),
+                        "label_path": str(labels_train / "invalid_000.txt"),
+                        "label_count": 1,
+                        "is_negative": False,
+                    },
+                    {
+                        "sample_slug": "missing",
+                        "sample_role": "background_candidate",
+                        "background_category": "sparse_building_context",
+                        "split": "val",
+                        "tile_index": 2,
+                        "kept": True,
+                        "image_path": str(image_val / "missing_000.png"),
+                        "label_path": str(missing_label_path),
+                        "label_count": 1,
+                        "is_negative": False,
+                    },
+                    {
+                        "sample_slug": "negative",
+                        "sample_role": "background_candidate",
+                        "background_category": "pure_empty_negative",
+                        "split": "val",
+                        "tile_index": 3,
+                        "kept": True,
+                        "image_path": str(image_val / "negative_000.png"),
+                        "label_path": str(labels_val / "negative_000.txt"),
+                        "label_count": 0,
+                        "is_negative": True,
+                    },
+                ],
+            }
+        ),
+        encoding="utf-8",
+    )
+
+    output_dir = tmp_path / "label-qa"
+    result = subprocess.run(
+        [
+            sys.executable,
+            str(script_path),
+            "--summary-path",
+            str(summary_path),
+            "--output-dir",
+            str(output_dir),
+            "--max-tiles",
+            "4",
+            "--columns",
+            "2",
+            "--thumb-size",
+            "128",
+        ],
+        cwd=ROOT,
+        check=True,
+        capture_output=True,
+        text=True,
+    )
+
+    assert "Operator YOLO label QA contact sheets rendered" in result.stdout
+
+    report = json.loads((output_dir / "operator_yolo_label_qa_summary.json").read_text(encoding="utf-8"))
+    assert report["status"] == "ok"
+    assert report["selected_tile_count"] == 4
+    assert report["rendered_tile_count"] == 4
+    assert report["missing_label_file_count"] == 1
+    assert report["invalid_label_count"] == 1
+    assert [tile["sample_slug"] for tile in report["selected_tiles"]] == [
+        "dense",
+        "invalid",
+        "missing",
+        "negative",
+    ]
+
+    sheet_path = output_dir / report["contact_sheets"][0]["path"]
+    assert sheet_path.exists()
+    sheet = Image.open(sheet_path).convert("RGB")
+    assert sheet.size[0] >= 256
+    assert sheet.size[1] >= 256
+    assert len(sheet.getcolors(maxcolors=1000000) or []) > 4
+
+    markdown = (output_dir / "operator_yolo_label_qa_contact_sheet.md").read_text(encoding="utf-8")
+    assert "Operator YOLO Label QA Contact Sheets" in markdown
+    assert "missing label files: 1" in markdown
+    assert "invalid label rows: 1" in markdown
+    assert "contact_sheet_001.png" in markdown