Warnings when exporting a fine tuned model from TensorFlow Object Detection API.

Prerequisites

Please answer the following questions for yourself before submitting an issue.

• [YES ] I am using the latest TensorFlow Model Garden release and TensorFlow 2.
• [YES ] I am reporting the issue to the correct repository. (Model Garden official or research directory)
• [YES ] I checked to make sure that this issue has not already been filed.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/tree/master/research/object_detection/exporter_main_v2.py
label_map_github_issue.txt
pipeline_github_issue.txt

2. Describe the bug

When exporting fine-tuned model (EfficientDet and MobileNet), I got Warnings / Info telling me:
WARNING:tensorflow:Skipping full serialization of Keras layer <object_detection.meta_architectures.ssd_meta_arch.SSDMetaArch object at 0x7f28c027c9e8>, because it is not built. W1121 16:00:38.569402 139816344586048 save_impl.py:78] Skipping full serialization of Keras layer <object_detection.meta_architectures.ssd_meta_arch.SSDMetaArch object at 0x7f28c027c9e8>, because it is not built.

Or:
INFO:tensorflow:Unsupported signature for serialization: (([(<tensorflow.python.framework.func_graph.UnknownArgument object at 0x7f27f8886dd8>, TensorSpec(shape=(None, 64, 64, 40), dtype=tf.float32, name='feature_pyramid/0/1')), (<tensorflow.python.framework.func_graph.UnknownArgument object at 0x7f27f8886e48>, TensorSpec(shape=(None, 32, 32, 112), dtype=tf.float32, name='feature_pyramid/1/1')), (<tensorflow.python.framework.func_graph.UnknownArgument object at 0x7f27f8886e10>, TensorSpec(shape=(None, 16, 16, 320), dtype=tf.float32, name='feature_pyramid/2/1'))], False), {}).

I really do not know their meaning. I got these errors in tf2.5 from source and tf2.3 binary.

warning_report_github.txt

3. Steps to reproduce

With trained model: from TensorFlow/models/research/: python object_detection/exporter_main_v2.py --input_type="image_tensor" --pipeline_config_path=$MODEL_DIR/pipeline.config --trained_checkpoint_dir=$MODEL_DIR --output_directory=$OUT

5. Additional context

See attached files.

After using exporter_main_v2.py, I tried to use https://github.com/opencv/opencv/tree/master/samples/dnn/tf_text_graph_efficientdet.py (I had to make some changes in order to ensure the compability with tf2). python path/to/tf_text_graph-efficientdet.py --input=./folder/saved_model.pb --output=./opencv_friendly/ --num_classes=1 it outputs:

2020-11-21 17:06:27.032604: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.1
Traceback (most recent call last):
  File "/path_to/dnn_utils/tf_text_graph_common.py", line 312, in write_text_graph
    cv.dnn.writeTextGraph(model_path, output_path)
cv2.error: OpenCV(4.4.0) /tmp/pip-req-build-hw4jq8lf/opencv/modules/dnn/src/tensorflow/tf_io.cpp:42: error: (-2:Unspecified error) FAILED: ReadProtoFromBinaryFile(param_file, param). Failed to parse GraphDef file: ./abo/frozen/saved_model/saved_model.pb in function 'ReadTFNetParamsFromBinaryFileOrDie'


During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "tf_text_graph_efficientdet.py", line 236, in <module>
    args.anchor_scale, args.num_classes, args.width, args.height)
  File "tf_text_graph_efficientdet.py", line 49, in createGraph
    write_text_graph(modelPath, outputPath, outNames)
  File "/home/neuro/PycharmProjects/Network/dnn_utils/tf_text_graph_common.py", line 319, in write_text_graph
    graph_def.ParseFromString(f.read())
google.protobuf.message.DecodeError: Error parsing message

6. System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 18.04
• Mobile device name if the issue happens on a mobile device: NA
• TensorFlow installed from (source or binary): Both
• TensorFlow version (use command below): 2.3 (binary) and 2.5 (source)
• Python version: 3.6
• Bazel version (if compiling from source): 3.7.0
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version: CUDA 11.1 cuDNN
• GPU model and memory: RTX 2700 8GB