One question on inception-v3 MXNet model

[handong1587]

First I want to thank your great job and sharing this comprehensive evaluation.
I am not clear if it is appropriate to post this question here but I am very curious that, in the inception-v3 MXNet page https://github.com/dmlc/mxnet-model-gallery/blob/master/imagenet-1k-inception-v3.md,
I downloaded its model file inception-v3.tar.gz, and I can see that in Inception-7-symbol.json, the "FullyConnected" op "fc1" has a "num_hidden" with 1008, which I thought should be 1000.
And actually @soeaver already convert it to a caffemodel with "fc1" layer with num_output of 1000.
I am new to MXNet and just want to know what happen to MXNet's "fc1", thanks ~

[soeaver]

@handong1587 Thank you for your attention.
As far as I am concerned, the MXNet version inception-v3 model is also converted from tensorflow. The original model has 1008 hidden layer neurons as classifier, bus it seems like the last 8 neurons are useless, so I remove them when I convert it to caffe with no loss of accuracy.

[handong1587]

@soeaver Got it. Thanks for your reply!