stu1130 commented on a change in pull request #12131: [MXNET-737][WIP] Add last 
batch handle for imageiter
URL: https://github.com/apache/incubator-mxnet/pull/12131#discussion_r210180174
 
 

 ##########
 File path: python/mxnet/image/image.py
 ##########
 @@ -1149,22 +1158,44 @@ def __init__(self, batch_size, data_shape, 
label_width=1,
         else:
             self.auglist = aug_list
         self.cur = 0
+        self._is_allowed_reading = True
+        self._cached_data = None
+        # handle the last batch
+        if self.seq and last_batch == 'discard':
+            new_seq_n = len(self.seq) - len(self.seq) % batch_size
+            self.seq = self.seq[:new_seq_n]
+
+        self.last_batch = last_batch
+        self.num_image = len(self.seq) if self.seq is not None else None
         self.reset()
 
     def reset(self):
         """Resets the iterator to the beginning of the data."""
-        if self.shuffle:
-            random.shuffle(self.seq)
+        if self.last_batch != 'roll_over' or \
+            self._is_allowed_reading is True:
+            if self.imgrec is not None:
+                self.imgrec.reset()
+            self.cur = 0
+        self._is_allowed_reading = True
 
 Review comment:
   We need to reset RecordIO, current cursor as long as the last_batch is not 
roll_over. As for self._is_allowed_reading, it needs to be set to True after 
reset. Consider the situation where the data iter with 'pad' or 'discard' have 
iterated till the last_batch, the self._is_allowed_reading is set to 
False(After did the padding, the value would set to False in case users call 
the next()), we need to set it back to True to prepare for next() function call.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to