Make FileOutcomeReport outcome labels configurable

lightly_studio/src/lightly_studio/core/file_outcome_report.py

@@ -94,9 +94,13 @@ class FileOutcomeReport:
     Attributes:
         max_examples_per_outcome: Maximum number of example paths kept per
             outcome.
+        label_overrides: Per-outcome display-label overrides used only by
+            `log_summary()`; any outcome absent from the dict falls back to
+            `outcome.value`. Defaults to empty (add-images behavior).
     """
 
     max_examples_per_outcome: int = DEFAULT_MAX_EXAMPLES_PER_OUTCOME
+    label_overrides: dict[FileOutcome, str] = field(default_factory=dict)
     _counts: dict[FileOutcome, int] = field(
         default_factory=lambda: dict.fromkeys(FileOutcome, 0),
         init=False,
@@ -171,11 +175,17 @@ def raise_if_all_failed(self) -> None:
                 f"({n_missing} missing, {n_broken} broken)."
             )
 
+    def _label(self, outcome: FileOutcome) -> str:
+        """Return the display label for `outcome`, honoring `label_overrides`."""
+        return self.label_overrides.get(outcome, outcome.value)
+
     def log_summary(self) -> None:
         """Log a single end-of-run summary of counts and example paths."""
-        counts = ", ".join(f"{outcome.value}={self._counts[outcome]}" for outcome in FileOutcome)
+        counts = ", ".join(
+            f"{self._label(outcome)}={self._counts[outcome]}" for outcome in FileOutcome
+        )
         logger.info(f"File outcomes: {counts}.")
         for outcome in FileOutcome:
             examples = self._example_paths[outcome]
             if examples:
-                logger.info(f"Example {outcome.value} paths: {', '.join(examples)}.")
+                logger.info(f"Example {self._label(outcome)} paths: {', '.join(examples)}.")

lightly_studio/src/lightly_studio/dataset/image_embedding.py

@@ -111,7 +111,7 @@ def load_and_preprocess(filepath: str) -> torch.Tensor | None:
     )
 
     kept_indices_set = set(result.kept_indices)
-    report = FileOutcomeReport()
+    report = FileOutcomeReport(label_overrides={FileOutcome.ADDED: "embedded"})
     for index, filepath in enumerate(filepaths):
         outcome = FileOutcome.ADDED if index in kept_indices_set else FileOutcome.BROKEN
         report.record(path=filepath, outcome=outcome)

lightly_studio/tests/core/test_file_outcome_report.py

@@ -140,3 +140,41 @@ def test_log_summary(self, caplog: pytest.LogCaptureFixture) -> None:
         assert "missing=1" in text
         assert "a.jpg" in text
         assert "b.jpg" in text
+        assert "Example added paths" in text
+
+    def test_log_summary__label_overrides(self, caplog: pytest.LogCaptureFixture) -> None:
+        report = FileOutcomeReport(label_overrides={FileOutcome.ADDED: "embedded"})
+        report.record(path="a.jpg", outcome=FileOutcome.ADDED)
+        report.record(path="b.jpg", outcome=FileOutcome.MISSING)
+
+        with caplog.at_level(logging.INFO):
+            report.log_summary()
+
+        text = caplog.text
+        # Overridden outcome uses the custom label.
+        assert "embedded=1" in text
+        assert "Example embedded paths" in text
+        assert "added=1" not in text
+        # Non-overridden outcomes still render their default labels.
+        assert "missing=1" in text
+
+    def test_log_summary__label_overrides_are_general(
+        self, caplog: pytest.LogCaptureFixture
+    ) -> None:
+        report = FileOutcomeReport(
+            label_overrides={
+                FileOutcome.ADDED: "embedded",
+                FileOutcome.MISSING: "not_found",
+            }
+        )
+        report.record(path="a.jpg", outcome=FileOutcome.ADDED)
+        report.record(path="b.jpg", outcome=FileOutcome.MISSING)
+
+        with caplog.at_level(logging.INFO):
+            report.log_summary()
+
+        text = caplog.text
+        assert "embedded=1" in text
+        assert "not_found=1" in text
+        assert "Example not_found paths" in text
+        assert "missing=1" not in text