thomelane commented on issue #11137: memory cost problem for different label width when using ssd example URL: https://github.com/apache/incubator-mxnet/issues/11137#issuecomment-404682592 Your `batch_size` is set to 12, so even if the total number of images in your dataset is smaller you're still inputting the same amount of data into the network in both cases. Saving in memory comes from having a smaller number of classes, but I wouldn't necessarily say this should be linear (i.e. memory footprint falling by the same amount as the labels). Most of the model parameters and activations will be from the feature maps at the start of the convolutional network, which is independent of the number of classes. You would only get a memory saving later on in the network.
---------------------------------------------------------------- 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: us...@infra.apache.org With regards, Apache Git Services