Emmanuel Pastor
Aug 25, 2017 · 1 min read

Hi Tran,

First of all, thank you very much for this very concise guide, so far it’s been tremendously helpful to ramp up the Tensorflow learning curve.

I’ve tried following everything and if I train my model from scratch (e.g. not from a checkpoint), everything seems to work well (except that I haven’t been able to train a model enough to actually detect anything), but if I try to start from a checkpoint as you recommend (I’ve tried both mobilenet and inception), I keep getting the following error message when I try to export the inference graph:

`InvalidArgumentError (see above for traceback): Assign requires shapes of both tensors to match. lhs shape= [3,3,128,12] rhs shape= [3,3,128,546]`

The weirdest part is that I’ve even tried exporting a vanilla checkpoint, freshly downloaded from Github and without doing any training of my own, and I still get that error (with different shape values, though). Any help would be tremendously appreciated! – thanks again.

)