knwng commented on issue #11883: ImageIter last batch / batch padding behavior URL: https://github.com/apache/incubator-mxnet/issues/11883#issuecomment-408743706 The data_batch and label_batch is initialized with a mx.nd.empty() object. Under my configuration, when the data generator is exhausted during filling a data batch, the data_batch will not be padded or filled with new data but remain garbage data even nan, which will raise an index out of bound error when calculate CrossEntropy metric. I'm currently change nd.empty into nd.zeros to initialize data_batch in the source code, but I think a automatically refilling or padding mechanism should be added to ImageIter.
---------------------------------------------------------------- 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
