Keras sequential model inputlayer issue

aswinjohn · July 6, 2018, 8:17pm

I have a sequential keras model with first layer as convolution2d, with input_shape defined. There is no explicit inputlayer.

When I tried to run it through tvm, it did not work I initially. Finally, what worked was

github.com

dmlc/tvm/blob/master/nnvm/python/nnvm/frontend/keras.py#L517


        for my_idx, node in enumerate(inbound_nodes):
            insym = []


            # Since Keras allows creating multiple layers from the same name instance,
            # we append node index to the symbol name to make it unique.
            # The one exception is InputLayer.  Changing input variable names after conversion
            # would confuse users, so we should keep them as far as possible.  Fortunately,
            # they are named uniquely to input_1, input_2, input_3 ... by default.
            for pred_idx, pred in zip(node.node_indices, node.inbound_layers):
                if isinstance(pred, keras.engine.InputLayer):
                    _sym = symtab.get_var(pred.name, must_contain=True)
                else:
                    _sym = symtab.get_var(pred.name + ':' + str(pred_idx), must_contain=True)
                insym.append(_sym)


            if len(insym) == 1:
                insym = insym[0]
            keras_op_to_nnvm(insym, keras_layer, keras_layer.name + ':' + str(my_idx), symtab)


outsym = symtab.get_var(model._output_layers[0].name + ':0')
tvmparams = {k:tvm.nd.array(np.array(v, dtype=np.float32)) for k, v in symtab.params.items()}

Changing ‘must_contain=True’ to ‘False’.
I think it happens because there is no inputlayer in model.layers, but the first conv2d still has predecessor as an implicit inputlayer.

Also, had to explicitly set the input_shape for the conv2d in keras.

Didn’t want to open an issue since it might be something I did not do correct.

Thanks,
Aswin .