[converter] Failed to convert keras model into ONNX format

[univerone]

Software and Hardware Versions

• ModelCI version latest
• CUDA Version v10.2
• GPU device used: true

Problem description

Failed to optimize ONNX model which is converted from a keras model

2021-02-28 16:25:40.416520: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2021-02-28 16:25:40.441655: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-02-28 16:25:40.442280: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: GeForce MX110 computeCapability: 5.0
coreClock: 1.006GHz coreCount: 2 deviceMemorySize: 1.96GiB deviceMemoryBandwidth: 37.33GiB/s
2021-02-28 16:25:40.442366: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory
2021-02-28 16:25:40.443859: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2021-02-28 16:25:40.445242: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2021-02-28 16:25:40.445510: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2021-02-28 16:25:40.447141: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2021-02-28 16:25:40.447996: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2021-02-28 16:25:40.448070: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudnn.so.7'; dlerror: libcudnn.so.7: cannot open shared object file: No such file or directory
2021-02-28 16:25:40.448079: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1598] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2021-02-28 16:25:40.448291: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2021-02-28 16:25:40.453816: I tensorflow/core/platform/profile_utils/cpu_utils.cc:102] CPU Frequency: 1800500000 Hz
2021-02-28 16:25:40.454180: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x7f5674000b60 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2021-02-28 16:25:40.454196: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2021-02-28 16:25:40.454257: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-02-28 16:25:40.454265: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108]      
tf executing eager_mode: True
tf.keras model eager_mode: False
The ONNX operator number change on the optimization: 458 -> 127
2021-02-28 16:25:54,773 - converter - INFO - Begin Simplify ONNX Model ...
2021-02-28 16:25:54,773 - converter - INFO - Begin Simplify ONNX Model ...
2021-02-28 16:25:54,773 - converter - INFO - Begin Simplify ONNX Model ...
2021-02-28 16:25:54,773 - converter - INFO - Begin Simplify ONNX Model ...
Traceback (most recent call last):
  File "~/ML-Model-CI/run_test.py", line 14, in <module>
    onnx_model = ONNXConverter.from_keras(keras_model)
  File "~/ML-Model-CI/modelci/hub/converter/onnx/converter.py", line 66, in wrap
    onnx_model = ONNXConverter.optim_onnx(onnx_model)
  File "~/ML-Model-CI/modelci/hub/converter/onnx/converter.py", line 224, in optim_onnx
    model = optimizer.optimize(model, passes)
  File "~/miniconda3/envs/modelci/lib/python3.8/site-packages/onnx/optimizer.py", line 55, in optimize
    optimized_model_str = C.optimize(model_str, passes)
IndexError: Input conv1_conv_W_new is undefined!

Process finished with exit code 1

Steps to Reproduce the Problem

from modelci.hub.converter.onnx import ONNXConverter
import tensorflow as tf

if __name__ == '__main__':
    keras_model = tf.keras.applications.ResNet50()
    onnx_model = ONNXConverter.from_keras(keras_model)

Expected Behavior

Other Information

Here is a revelant issue onnx/keras-onnx#337
Maybe we should remove optimize process after the conversion of keras model to onnx model