chore: Add reading_order evaluation and visualization in the examples…

… for dp-bench and omnidocbench Add the doc_id in the evaluation report Signed-off-by: Nikos Livathinos <[email protected]>
DS4SD · Jan 14, 2025 · 96a7d88 · 96a7d88
1 parent 36b0722
commit 96a7d88
Show file tree

Hide file tree

Showing 3 changed files with 27 additions and 0 deletions.
diff --git a/docling_eval/evaluators/readingorder_evaluator.py b/docling_eval/evaluators/readingorder_evaluator.py
@@ -19,6 +19,8 @@
 
 
 class PageReadingOrderEvaluation(BaseModel):
+    doc_id: str
+
     # BBoxes are in BOTTOMLEFT origin and in the true order
     bboxes: List[Tuple[float, float, float, float]]
     pred_order: List[int]
@@ -75,6 +77,7 @@ def __call__(
             ards.append(ard)
 
             page_evaluation = PageReadingOrderEvaluation(
+                doc_id=doc_id,
                 bboxes=[b.as_tuple() for b in reading_order["bboxes"]],
                 pred_order=reading_order["pred_order"],
                 ard=ard,

diff --git a/docs/examples/benchmark_dpbench.py b/docs/examples/benchmark_dpbench.py
@@ -58,6 +58,18 @@ def main():
             idir=odir_lay,
             odir=odir_lay,
         )
+        evaluate(
+            modality=EvaluationModality.READING_ORDER,
+            benchmark=BenchMarkNames.DPBENCH,
+            idir=odir_lay,
+            odir=odir_lay,
+        )
+        visualise(
+            modality=EvaluationModality.READING_ORDER,
+            benchmark=BenchMarkNames.DPBENCH,
+            idir=odir_lay,
+            odir=odir_lay,
+        )
 
     if True:
         create_dpbench_tableformer_dataset(

diff --git a/docs/examples/benchmark_omnidocbench.py b/docs/examples/benchmark_omnidocbench.py
@@ -59,6 +59,18 @@ def main():
             idir=odir_lay,
             odir=odir_lay,
         )
+        evaluate(
+            modality=EvaluationModality.READING_ORDER,
+            benchmark=BenchMarkNames.OMNIDOCBENCH,
+            idir=odir_lay,
+            odir=odir_lay,
+        )
+        visualise(
+            modality=EvaluationModality.READING_ORDER,
+            benchmark=BenchMarkNames.OMNIDOCBENCH,
+            idir=odir_lay,
+            odir=odir_lay,
+        )
 
     if True:
         create_omnidocbench_tableformer_dataset(