diff --git a/segment_eval_inference.py b/segment_eval_inference.py
index 4676ed62..c5813aa6 100644
--- a/segment_eval_inference.py
+++ b/segment_eval_inference.py
@@ -46,7 +46,7 @@ def labels(self):
 
   def to_file(self, file_name):
     """Materialize the GT mapping to file."""
-    with tf.gfile.Open(file_name, "w") as fobj:
+    with tf.io.gfile.GFile(file_name, "w") as fobj:
       for k, v in self._labels.items():
         seg_id, label = k
         line = "%s,%s,%s\n" % (seg_id, label, v)
@@ -79,18 +79,18 @@ def read_labels(data_pattern, cache_path=""):
       tf.logging.info("Reading cached labels from %s..." % cache_path)
       return Labels.from_file(cache_path)
   tf.enable_eager_execution()
-  data_paths = tf.gfile.Glob(data_pattern)
+  data_paths = tf.io.gfile.glob(data_pattern)
   ds = tf.data.TFRecordDataset(data_paths, num_parallel_reads=50)
   context_features = {
-      "id": tf.FixedLenFeature([], tf.string),
-      "segment_labels": tf.VarLenFeature(tf.int64),
-      "segment_start_times": tf.VarLenFeature(tf.int64),
-      "segment_scores": tf.VarLenFeature(tf.float32)
+      "id": tf.io.FixedLenFeature([], tf.string),
+      "segment_labels": tf.io.VarLenFeature(tf.int64),
+      "segment_start_times": tf.io.VarLenFeature(tf.int64),
+      "segment_scores": tf.io.VarLenFeature(tf.float32)
   }
 
   def _parse_se_func(sequence_example):
-    return tf.parse_single_sequence_example(sequence_example,
-                                            context_features=context_features)
+    return tf.io.parse_single_sequence_example(
+        sequence_example, context_features=context_features)
 
   ds = ds.map(_parse_se_func)
   rated_labels = {}
@@ -104,7 +104,8 @@ def _parse_se_func(sequence_example):
     segment_scores = cxt_feature_val["segment_scores"].values.numpy()
     for label, start_time, score in zip(segment_labels, segment_start_times,
                                         segment_scores):
-      rated_labels[("%s:%d" % (video_id, start_time), label)] = score
+      rated_labels[("%s:%d" % (video_id.decode("utf8"), start_time),
+                    label)] = score
     batch_id = len(rated_labels) // batch_size
     if batch_id != last_batch:
       tf.logging.info("%d examples processed.", len(rated_labels))
@@ -129,10 +130,13 @@ def read_segment_predictions(file_path, labels, top_n=None):
     a segment prediction list for each classes.
   """
   cls_preds = {}  # A label_id to pred list mapping.
-  with tf.gfile.Open(file_path) as fobj:
+  with tf.io.gfile.GFile(file_path) as fobj:
     tf.logging.info("Reading predictions from %s..." % file_path)
     for line in fobj:
       label_id, pred_ids_val = line.split(",")
+      if not label_id.isdigit():
+        # Skip the header line.
+        continue
       pred_ids = pred_ids_val.split(" ")
       if top_n:
         pred_ids = pred_ids[:top_n]
@@ -177,9 +181,11 @@ def main(unused_argv):
           float(x) / len(class_preds) for x in range(len(class_preds), 0, -1)
       ]
     seg_scored_preds.append(seg_scored_pred)
-    num_positives.append(positive_counter[label_id])
+    num_positives.append(positive_counter.get(label_id, 0))
+
   map_cal.accumulate(seg_scored_preds, seg_labels, num_positives)
-  map_at_n = np.mean(map_cal.peek_map_at_n())
+  aps = map_cal.peek_map_at_n()
+  map_at_n = np.mean(aps)
   tf.logging.info("Num classes: %d | mAP@%d: %.6f" %
                   (len(seg_preds), FLAGS.top_n, map_at_n))