diff --git a/src/transformers/onnx/config.py b/src/transformers/onnx/config.py
index fdcc12bdcd1f0e..0dbbaba4a5047c 100644
--- a/src/transformers/onnx/config.py
+++ b/src/transformers/onnx/config.py
@@ -99,6 +99,7 @@ class OnnxConfig(ABC):
                 "end_logits": {0: "batch", 1: "sequence"},
             }
         ),
+        "semantic-segmentation": OrderedDict({"logits": {0: "batch", 1: "sequence"}}),
         "seq2seq-lm": OrderedDict({"logits": {0: "batch", 1: "decoder_sequence"}}),
         "sequence-classification": OrderedDict({"logits": {0: "batch"}}),
         "token-classification": OrderedDict({"logits": {0: "batch", 1: "sequence"}}),
diff --git a/src/transformers/onnx/features.py b/src/transformers/onnx/features.py
index fbfeb47250e73f..841938d01a0200 100644
--- a/src/transformers/onnx/features.py
+++ b/src/transformers/onnx/features.py
@@ -25,6 +25,7 @@
         AutoModelForMultipleChoice,
         AutoModelForObjectDetection,
         AutoModelForQuestionAnswering,
+        AutoModelForSemanticSegmentation,
         AutoModelForSeq2SeqLM,
         AutoModelForSequenceClassification,
         AutoModelForTokenClassification,
@@ -36,6 +37,7 @@
         TFAutoModelForMaskedLM,
         TFAutoModelForMultipleChoice,
         TFAutoModelForQuestionAnswering,
+        TFAutoModelForSemanticSegmentation,
         TFAutoModelForSeq2SeqLM,
         TFAutoModelForSequenceClassification,
         TFAutoModelForTokenClassification,
@@ -94,6 +96,7 @@ class FeaturesManager:
             "image-classification": AutoModelForImageClassification,
             "image-segmentation": AutoModelForImageSegmentation,
             "masked-im": AutoModelForMaskedImageModeling,
+            "semantic-segmentation": AutoModelForSemanticSegmentation,
         }
     if is_tf_available():
         _TASKS_TO_TF_AUTOMODELS = {
@@ -105,6 +108,7 @@ class FeaturesManager:
             "token-classification": TFAutoModelForTokenClassification,
             "multiple-choice": TFAutoModelForMultipleChoice,
             "question-answering": TFAutoModelForQuestionAnswering,
+            "semantic-segmentation": TFAutoModelForSemanticSegmentation,
         }
 
     # Set of model topologies we support associated to the features supported by each topology and the factory
@@ -236,7 +240,7 @@ class FeaturesManager:
         "data2vec-vision": supported_features_mapping(
             "default",
             "image-classification",
-            "image-segmentation",
+            "semantic-segmentation",
             onnx_config_cls="models.data2vec.Data2VecVisionOnnxConfig",
         ),
         "deberta": supported_features_mapping(
diff --git a/tests/onnx/test_onnx_v2.py b/tests/onnx/test_onnx_v2.py
index 5634abc7706856..ac18a9463b1673 100644
--- a/tests/onnx/test_onnx_v2.py
+++ b/tests/onnx/test_onnx_v2.py
@@ -284,6 +284,13 @@ def _onnx_export(self, test_name, name, model_name, feature, onnx_config_class_c
         model_class = FeaturesManager.get_model_class_for_feature(feature)
         config = AutoConfig.from_pretrained(model_name)
         model = model_class.from_config(config)
+
+        # It seems `nn.AdaptiveAvgPool2d` with `output_size` > 1 is not working with ONNX. We get `Segmentation fault`
+        # without any useful information.
+        if model.__class__.__name__ == "Data2VecVisionForSemanticSegmentation":
+            config.pool_scales = [1] * len(config.pool_scales)
+            model = model_class.from_config(config)
+
         onnx_config = onnx_config_class_constructor(model.config)
 
         if is_torch_available():