lrcn model to onnx error

[Magsun]

@harvitronix hello Harvey, thanks to your really good tutorial for video classification.

I followed your instruction to train my own lrcn video classification model and it works well under tensorflow&keras. But when I tried to convert this .hdf5 model to .onnx, I met this error below:

env info:
tf-gpu 1.14
tb-nightly
cuda10.1
onnx 1.7.0
keras2onnx 1.6.1
python3.6

Traceback (most recent call last):
File "/workspace/h52onnx.py", line 15, in
onnx_model = keras2onnx.convert_keras(model, model.name)
File "/opt/conda/lib/python3.6/site-packages/keras2onnx/main.py", line 95, in convert_keras
return convert_topology(topology, name, doc_string, target_opset, channel_first_inputs)
File "/opt/conda/lib/python3.6/site-packages/keras2onnx/topology.py", line 243, in convert_topology
get_converter(operator.type)(scope, operator, container)
File "/opt/conda/lib/python3.6/site-packages/keras2onnx/ke2onnx/main.py", line 94, in convert_keras_flatten
shape_len = len(iop.input_shape)
File "/opt/conda/lib/python3.6/site-packages/keras/engine/base_layer.py", line 918, in input_shape
raise AttributeError('The layer has never been called '
AttributeError: The layer has never been called and thus has no defined input shape.

For your information, I only changed input size from self.input_shape = (seq_length, 80, 80, 3) to self.input_shape = (seq_length, 320, 320, 3).

Convert code listed below:

import keras
import keras2onnx
import onnx
from keras.models import load_model

model = load_model('/workspace/lrcn-images.050-1.443.hdf5')
onnx_model = keras2onnx.convert_keras(model, model.name)
temp_model_file = '/workspace/model.onnx'
onnx.save_model(onnx_model, temp_model_file)

Kindly help. Thanks again.