transform_param not working in HDF5_DATA layer -> Anybody knows a workaround?

[Armin Kappeler]

I noticed that the transform_param features don't work for the HDF5 data layer. 

I'd like to use crop_size and mirror. Is there a workaround for this? 

My input data has 6 channels (dimension=120 x 120 x 6 x batchsize), otherwise I'd just use an image_data layer.

Also, how do I let the caffe-developers know? they might want to know that, too (if they don't already)

Thanks

Armin

[Ali Mollahosseini]

I am not sure but I think is  transform_param only belongs to 

  type: "Data"

layers, not HDF5. 

In this case, better to write your own convert_imageset that concatenate images in the blob of 6 channels and save them as LMDB or leveldb. Then you can use slice layer to separate them in your model.