Redesigned annotators

Linezonetrigger

[resolve #1] Review comment by @copilot (PR #2185):
"isinstance(..., int) does not assert iscrowd isn't a bool because bool is a subclass of int"

- Replace hardcoded `cv2.mean(roi)[:3]` with ndim-aware fill: scalar for grayscale, channel-matched tuple for colour images; avoids shape mismatch broadcast error on single-channel frames

[resolve roboflow#1] Review comment by @Copilot (PR roboflow#709):
"If scene/roi is a single-channel (grayscale) array, this will fail due to shape mismatch..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

- Break error message string at line 686 to stay within 88-char ruff E501 limit
- Add explicit npt.NDArray[np.float32] | None annotation on conf_selected; use
  cast() at the branch assignment to prevent mypy widening to Optional[Any]
  (mypy 1.19.1 + numpy 2.x stubs treat npt.NDArray as Any, so narrowing via
  `if conf_selected is not None` requires a concrete type from cast, and the
  self.confidence access needs its own is-not-None guard inside the loop)

[resolve #1] /review finding by sw-engineer (report: _outputs/2026/03/output-review-2026-03-30-5.md):
"conf_selected inferred as Optional[Any]; mypy pre-commit hook fails"
[resolve #3] /review finding by linting-expert (report: _outputs/2026/03/output-review-2026-03-30-5.md):
"ruff E501: line too long (89 > 88) at core.py:686"

---
Co-authored-by: Claude Code <noreply@anthropic.com>

- [resolve #1] RUF015: replace list(annotations.values())[0] with next(iter(annotations.values()))
- [resolve #2] E501: split assertion message to stay within 88-char limit
- [resolve #3] ruff format: committed formatted diff for test file

---
Co-authored-by: Claude Code <noreply@anthropic.com>

- Add test_yolo_polygon_mask_precision_no_coord_drift verifying that
  load → save round-trip keeps coordinate drift under 0.005 normalised units

[resolve roboflow#1] Review comment by @Borda (PR roboflow#1746):
"could you pls add a test to demonstrate the previous issue and that your fix resolves it"

- Add test_polygons_to_masks_multiple_polygons_shape: verifies (N, H, W)
  output shape with fractional pixel coords and disjoint mask assertion
- Add test_yolo_polygon_mask_precision_no_coord_drift: round-trip load/save
  with odd resolution (101x97) to ensure non-integer pixel coordinates

[resolve roboflow#1] Review comment by @Borda (PR roboflow#1746):
"could you pls add a test to demonstrate the previous issue and that your fix resolves it"

- Remove frame_rate argument from ByteTrack() in smoother.py docstring
- Remove frame_rate argument from 4 example scripts (speed_estimation, heatmap_and_track)
- heatmap_and_track already passes lost_track_buffer explicitly; frame_rate=fps was redundant

[resolve #1] Review comment by @Copilot (gh): `frame_rate` is removed from `__init__`, but there are still in-repo call sites that pass `frame_rate=`...

---
Co-authored-by: Claude Code <noreply@anthropic.com>

The else branch added by the original fix used hasattr(value, "__getitem__")
to decide whether to slice custom_data values per detection. This incorrectly
indexes dicts by integer 0, raising KeyError when custom_data contains dict
values (e.g. {"metadata": {"sensor_id": 101}}).

Non-ndarray types (dicts, scalars, lists used as a single value per detection)
should be written as-is. Only np.ndarray values require per-row indexing.

[resolve roboflow#1] /review finding by sw-engineer (report: .temp/output-review-fix-csv-json-sink-custom-data-array-slicing-2026-04-13.md): CSVSink else branch KeyError:0 on dict custom_data at csv_sink.py:153

---
Co-authored-by: Claude Code <noreply@anthropic.com>

[resolve #1] Review comment by @Copilot (gh):
"mask from rle_to_mask is a boolean array, but masks is annotated as list[npt.NDArray[np.uint8]]..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

…ings

- Describe broadcast vs. per-row slicing rules for custom_data values in
  CSVSink.append() and JSONSink.append()

[resolve roboflow#1] /review finding by foundry:sw-engineer + foundry:doc-scribe (report: .temp/output-review-fix-sinks-slice-list-values-2026-04-17.md)

---
Co-authored-by: Claude Code <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <codex@openai.com>

- /0.*/ block was misplaced after GoogleOther group (blank line ends a
  group in robots.txt), making it a dangling directive most parsers
  would ignore; move it into User-agent:* so all crawlers respect it

[resolve #1] Review comment by @Copilot (PR #2224):
"Disallow: /0.*/ directive is currently under the User-agent: GoogleOther
group, so it only applies to that crawler..."

---
Co-authored-by: Claude Code <noreply@anthropic.com>

[resolve #1] Review comment by @Copilot (PR #2232):
"The cookbook card title has a spelling error (\"Underestand\"). This is user-facing text and will be indexed/visible in the docs UI; please correct the display name while keeping the existing slug if that path is intentional."

- Fixed heading in docs/notebooks/underestand-visitors-with-yolo-world.ipynb
- Slug/filename preserved as Copilot instructed

---
Co-authored-by: Claude Code <noreply@anthropic.com>

…r (report: .reports/review/2026-05-19T10-05-17Z/review-report.md): add timeout=300 to subprocess.run

Without a timeout, ffmpeg can hang indefinitely on malformed containers or
stalled filesystems, silently freezing the caller's program. The existing
`except Exception` block already catches `subprocess.TimeoutExpired`, so
adding `timeout=300` is the minimal fix — it logs a warning and returns.

300 s is a conservative cap sufficient for typical video remux workloads.

---
Co-authored-by: Claude Code <noreply@anthropic.com>

[resolve #1] Review by @Copilot (PR #2260):
"_extract_class_names() removed isinstance(data, dict) guard; can raise AttributeError on malformed YAML"
Challenge: evidence=VALID (self-verified) suggestion=VALID resolution=as-suggested

- Restore isinstance(data, dict) check before .get("names") in yolo.py
- Fix pre-existing [no-any-return] mypy error in annotators/core.py:_load_icon

---
Co-authored-by: Claude Code <noreply@anthropic.com>

Relocates test_from_coco_multi_segment_masks out of TestClassNamePopulation
(scoped to CLASS_NAME_DATA_FIELD propagation) into a new TestFromCocoMasks
class in tests/dataset/formats/test_coco.py.

- Adds TestFromCocoMasks class + coco_data_with_multi_segment_segmentation
  fixture in tests/dataset/formats/test_coco.py
- Test now calls DetectionDataset.from_coco without force_masks (default
  False) — exercises _with_seg_mask auto-detection, the natural user path
- Removes fixture, test, and unused `import json` from tests/dataset/test_core.py

Resolves [resolve #1] @copilot-pull-request-reviewer (gh):
move test out of TestClassNamePopulation

---
Co-authored-by: Claude Code <noreply@anthropic.com>

…fault to 0.0

Pre-existing default of 0.75 caused aggressive polygon simplification for direct
callers while as_coco used 0.0 — inconsistent behavior for the same param.

[resolve roboflow#1] /review finding by foundry:solution-architect (report: .reports/review/2026-05-22T18-44-11Z/review-report.md):

---
Co-authored-by: Claude Code <noreply@anthropic.com>

…roboflow_result

Update three parametrize entries to use np.empty(0, dtype=str) instead of np.empty(0)
for expected class_name on empty-predictions paths. Add explicit dtype equality assertion
alongside the existing np.array_equal check so future regressions at the helper layer
are caught (np.array_equal is dtype-blind on zero-length arrays).

[resolve roboflow#1] /review finding by foundry:qa-specialist (report: .reports/review/2026-05-27T06-06-27Z/review-report.md)

---
Co-authored-by: Claude Code <noreply@anthropic.com>

[resolve #1] Review by @Copilot (PR #2294):
"_validate_keypoint_confidence validates a 2D array with shape (n, m),
but expected_shape uses f"({n, m})" (set literal) and message says 1D"
Challenge: evidence=VALID suggestion=VALID resolution=as-suggested

- Fix f"({n, m})" -> f"({n}, {m})" (set literal -> proper shape string)
- Fix error message "1D" -> "2D" to match actual array dimensionality

---
Co-authored-by: Claude Code <noreply@anthropic.com>