diff --git a/mediapipe/util/sequence/media_sequence.cc b/mediapipe/util/sequence/media_sequence.cc
index 62bfa19b4..f76c53295 100644
--- a/mediapipe/util/sequence/media_sequence.cc
+++ b/mediapipe/util/sequence/media_sequence.cc
@@ -160,6 +160,10 @@ absl::Status ReconcileMetadataFeatureFloats(
     const std::string& key = key_value.first;
     if (absl::StrContains(key, kFeatureFloatsKey)) {
       const auto prefix = key.substr(0, key.find(kFeatureFloatsKey) - 1);
+      if (GetFeatureFloatsSize(prefix, *sequence) < 1) {
+        // Unable to determine the feature dimensions as no data is provided.
+        continue;
+      }
       int number_of_elements = GetFeatureFloatsAt(prefix, *sequence, 0).size();
       if (HasFeatureDimensions(prefix, *sequence) &&
           !GetFeatureDimensions(prefix, *sequence).empty()) {
diff --git a/mediapipe/util/sequence/media_sequence_test.cc b/mediapipe/util/sequence/media_sequence_test.cc
index 42b0e3889..0797ed472 100644
--- a/mediapipe/util/sequence/media_sequence_test.cc
+++ b/mediapipe/util/sequence/media_sequence_test.cc
@@ -23,6 +23,7 @@
 #include "mediapipe/framework/port/opencv_imgcodecs_inc.h"
 #include "mediapipe/framework/port/status_matchers.h"
 #include "tensorflow/core/example/example.pb.h"
+#include "tensorflow/core/example/feature.pb.h"
 
 namespace mediapipe {
 namespace mediasequence {
@@ -867,6 +868,8 @@ TEST(MediaSequenceTest, ReconcileMetadataFloats) {
   AddFeatureFloats(feature_name, vf, &sequence);
   AddFeatureTimestamp(feature_name, 1000000, &sequence);
   AddFeatureTimestamp(feature_name, 2000000, &sequence);
+  sequence.mutable_feature_lists()->mutable_feature_list()->insert(
+      {"EMPTY/feature/floats", tensorflow::FeatureList()});
 
   MP_ASSERT_OK(ReconcileMetadata(true, false, &sequence));
   ASSERT_EQ(GetFeatureDimensions(feature_name, sequence).size(), 1);