changes w.r.t review

src/transformers/models/eomt/image_processing_eomt.py

@@ -858,6 +858,7 @@ class EomtImageProcessor(BaseImageProcessor):
     ):
         """Post-processes model outputs into final panoptic segmentation prediction."""
 
+        # `mask_threshold` and `overlap_mask_area_threshold` args are unused and only present for Pipeline compatability.
         size = size if size is not None else self.size
 
         masks_queries_logits = outputs.masks_queries_logits  # [batch_size, num_queries, height, width]

src/transformers/models/eomt/image_processing_eomt_fast.py

@@ -522,6 +522,7 @@ class EomtImageProcessorFast(BaseImageProcessorFast):
     ):
         """Post-processes model outputs into Instance Segmentation Predictions."""
 
+        # `mask_threshold` and `overlap_mask_area_threshold` args are unused and only present for Pipeline compatability.
         size = size if size is not None else self.size
 
         masks_queries_logits = outputs.masks_queries_logits

tests/models/eomt/test_modeling_eomt.py

@@ -474,4 +474,9 @@ class EomtForUniversalSegmentationIntegrationTest(unittest.TestCase):
         image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
 
         pipe = pipeline(model=self.model_id, subtask="panoptic", device=torch_device)
-        _ = pipe(image)
+        output = pipe(image)
+
+        EXPECTED_OUTPUT_LABELS = ["LABEL_15", "LABEL_15", "LABEL_57", "LABEL_65", "LABEL_65"]
+
+        output_labels = [segment["label"] for segment in output["segments_info"]]
+        self.assertEqual(output_labels, EXPECTED_OUTPUT_LABELS)