nervetumer
nervetumer
The ArrayIterator needs to have the data as a 2D matrix with the dim (N, CHW). So the CHW part needs to be squashed and it needs the shape parameter...
Could you pull the most recent model zoo repo and try again, I think the model needed to be updated to use the SingleOutputTree
Not sure if this relates to #3. Would be good to make sure that this is working for you.
This gist [link](https://gist.github.com/nervetumer/6c5777f31f0951bb2a1a54b10d9e3b42) shows how to transfer the trained alexnet model over to a network with only 10 outputs in the last fully connected layer. It may give enough information...