Does channel prune rmt support nchw model?

[ChaoYu123]

I use a NCHW model as input for channel_pruning_rmt and change the corresponding index of tensors. However, I fail at "assert err_full < 1e-6, 'unable to recover output feature maps - full (%e)' % err_full" in __smpl_inputs_n_outputs function. I wonder if channel pruning rmt supports NCHW model for channel pruning.