Dataset Tests

docs/.gitignore

@@ -3,4 +3,5 @@ Gemfile.lock
 .DS_store
 .jekyll-cache
 .gem
-.bundle
+.bundle
+vendor/

docs/_pages/gui.md

@@ -28,6 +28,22 @@ The GUI will open in your default web browser, typically at `http://localhost:85
 
 The GUI consists of three main tabs, each designed for specific tasks:
 
+### 🎥 GUI Tutorial Videos
+
+The following videos show the GUI workflow for the currently supported tasks:
+
+#### Image Detection GUI
+
+<iframe src="https://drive.google.com/file/d/1iNk-V3rwN9tZ2pAd_HONAkVLm3JstF00/preview" width="100%" height="420" allow="autoplay" allowfullscreen></iframe>
+
+#### Image Segmentation GUI
+
+<iframe src="https://drive.google.com/file/d/16ixtrevvqDn6l2iuYXxHBcm4Z5IwSlOl/preview" width="100%" height="420" allow="autoplay" allowfullscreen></iframe>
+
+#### LiDAR Segmentation GUI
+
+<iframe src="https://drive.google.com/file/d/1BDT6fjLFPlG3HuHE9wWhCxwKQJ_X2PjQ/preview" width="100%" height="420" allow="autoplay" allowfullscreen></iframe>
+
 ### 📁 Dataset Viewer Tab
 
 The Dataset Viewer allows you to explore and visualize your datasets before running evaluations.

tests/test_cityscapes.py

@@ -0,0 +1,161 @@
+from typing import Any, Dict, List, Tuple
+from unittest.mock import patch
+
+import numpy as np
+from PIL import Image
+import pandas as pd
+import pytest
+
+from perceptionmetrics.datasets.cityscapes import (
+    CityscapesImageSegmentationDataset,
+    build_dataset,
+    build_dataset_ontology,
+    build_train_id_ontology_translation,
+)
+
+_FAKE_ROOT = "/fake/cityscapes"
+_FAKE_IMAGE = (
+    "/fake/cityscapes/leftImg8bit_trainvaltest/leftImg8bit/val/"
+    "frankfurt/frankfurt_000000_000294_leftImg8bit.png"
+)
+_FAKE_LABEL = (
+    "/fake/cityscapes/gtFine/val/frankfurt/"
+    "frankfurt_000000_000294_gtFine_labelIds.png"
+)
+
+
+def _make_patched_build_dataset(
+    image_files_by_split: Dict[str, List[str]], label_exists: bool = True
+) -> Tuple[dict, dict]:
+    """Return a call to build_dataset with filesystem calls mocked.
+
+    :param image_files_by_split: Mapping of split name to list of image file paths
+    :type image_files_by_split: dict
+    :param label_exists: Whether the expected label path should exist
+    :type label_exists: bool
+    :return: Result of build_dataset
+    :rtype: tuple
+    """
+
+    def _fake_glob(pattern: str) -> List[str]:
+        for split, files in image_files_by_split.items():
+            if f"/{split}/" in pattern:
+                return files
+        return []
+
+    def _fake_exists(path: str) -> bool:
+        return label_exists and path == _FAKE_LABEL
+
+    with patch("perceptionmetrics.datasets.cityscapes.glob", side_effect=_fake_glob):
+        with patch(
+            "perceptionmetrics.datasets.cityscapes.os.path.exists",
+            side_effect=_fake_exists,
+        ):
+            return build_dataset(val_dataset_root=_FAKE_ROOT)
+
+
+def test_build_dataset() -> None:
+    """Regression tests for Cityscapes build_dataset and ontology handling.
+
+    Verifies that:
+    - Cityscapes images are matched with their expected label files.
+    - Missing labels are skipped.
+    - Raw label IDs are used by default.
+    - Train IDs are used when requested.
+    - Train-ID mode requires the train-ID label suffix.
+    - Ontology translation maps valid classes to themselves.
+    """
+    dataset, ontology = _make_patched_build_dataset({"val": [_FAKE_IMAGE]})
+
+    assert isinstance(dataset, dict)
+    assert "frankfurt_000000_000294" in dataset
+    assert dataset["frankfurt_000000_000294"] == (
+        _FAKE_IMAGE,
+        _FAKE_LABEL,
+        "frankfurt",
+        "val",
+    )
+    assert "road" in ontology
+    assert ontology["road"]["idx"] == ontology["road"]["cityscapes_id"]
+
+    dataset, _ = _make_patched_build_dataset(
+        {"val": [_FAKE_IMAGE]},
+        label_exists=False,
+    )
+    assert dataset == {}
+
+    train_id_ontology = build_dataset_ontology(use_train_id=True)
+    assert train_id_ontology["road"]["idx"] == train_id_ontology["road"]["train_id"]
+    assert ontology["road"]["idx"] != train_id_ontology["road"]["idx"]
+
+    with pytest.raises(ValueError, match="use_train_id=True requires train-id labels"):
+        build_dataset(val_dataset_root=_FAKE_ROOT, use_train_id=True)
+
+    translation = build_train_id_ontology_translation()
+    assert translation["road"] == "road"
+    assert translation["car"] == "car"
+
+
+def test_cityscapes_dataset() -> None:
+    """Regression tests for the Cityscapes dataset wrapper.
+
+    Verifies that:
+    - The wrapper converts the built dataset into a DataFrame.
+    - Empty datasets raise a clear error.
+    """
+    fake_dataset = {
+        "frankfurt_000000_000294": (
+            _FAKE_IMAGE,
+            _FAKE_LABEL,
+            "frankfurt",
+            "val",
+        )
+    }
+    fake_ontology: Dict[str, Any] = {"road": {"idx": 7, "rgb": [128, 64, 128]}}
+
+    with patch(
+        "perceptionmetrics.datasets.cityscapes.build_dataset",
+        return_value=(fake_dataset, fake_ontology),
+    ):
+        dataset = CityscapesImageSegmentationDataset(val_dataset_root=_FAKE_ROOT)
+
+    assert isinstance(dataset.dataset, pd.DataFrame)
+    assert len(dataset.dataset) == 1
+    assert dataset.dataset.index.tolist() == ["frankfurt_000000_000294"]
+    assert dataset.dataset.loc["frankfurt_000000_000294", "scene"] == "frankfurt"
+    assert dataset.dataset.loc["frankfurt_000000_000294", "split"] == "val"
+    assert dataset.dataset_dir == _FAKE_ROOT
+
+    with patch(
+        "perceptionmetrics.datasets.cityscapes.build_dataset",
+        return_value=({}, fake_ontology),
+    ):
+        with pytest.raises(ValueError, match="No Cityscapes samples were found"):
+            CityscapesImageSegmentationDataset(val_dataset_root=_FAKE_ROOT)
+
+
+def test_cityscapes_read_label(tmp_path) -> None:
+    """Verify that Cityscapes label-ID images are read as label arrays."""
+    label_path = tmp_path / "frankfurt_000000_000294_gtFine_labelIds.png"
+    expected = np.array([[7, 8, 11], [12, 13, 17]], dtype=np.uint8)
+    Image.fromarray(expected).save(label_path)
+
+    fake_dataset = {
+        "frankfurt_000000_000294": (
+            _FAKE_IMAGE,
+            str(label_path),
+            "frankfurt",
+            "val",
+        )
+    }
+    fake_ontology: Dict[str, Any] = {"road": {"idx": 7, "rgb": [128, 64, 128]}}
+
+    with patch(
+        "perceptionmetrics.datasets.cityscapes.build_dataset",
+        return_value=(fake_dataset, fake_ontology),
+    ):
+        dataset = CityscapesImageSegmentationDataset(val_dataset_root=_FAKE_ROOT)
+
+    label = dataset.read_label(str(label_path))
+
+    assert np.array_equal(label, expected)