Which config should I use to train model for Android (covert later to .tflite format)?

[anonym24]

Which config should I use to train model for Android (covert later to .tflite format)?

Tensorflow has TOCO converter (.pb > .tflite)

But what config should I use to train pb file?
ssdlite_mobilenet_v1_coco.config, ssd_mobilenet_v1_300x300_coco14_sync.config or ssd_mobilenet_v1_quantized_300x300_coco14_sync.config?

[anonym24]

Also there is ssd_mobilenet_v2_quantized_coco pretrained model available, but no config for this one

[anonym24]

when using python train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/ssd_mobilenet_v1_quantized_300x300_coco14_sync.config

ERROR:tensorflow:==================================
Object was never used (type <class 'tensorflow.python.framework.ops.Tensor'>):

[anonym24]

I also tried this project on Ubuntu, faster_rcnn_inception_v2_pets.config works fine (the same for Windows)

but ssd_mobilenet_v1_quantized_300x300_coco14_sync.config fails on both Windows and Ubuntu

Ubuntu error when using ssd_mobilenet_v1_quantized_300x300_coco14_sync.config:

python3 train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/ssd_mobilenet_v1_300x300_coco14_sync.config
WARNING:tensorflow:From /home/user/.local/lib/python3.6/site-packages/tensorflow/python/platform/app.py:125: main (from __main__) is deprecated and will be removed in a future version.
Instructions for updating:
Use object_detection/model_main.py.
W1107 14:10:17.052690 140440119306048 tf_logging.py:125] From /home/user/.local/lib/python3.6/site-packages/tensorflow/python/platform/app.py:125: main (from __main__) is deprecated and will be removed in a future version.
Instructions for updating:
Use object_detection/model_main.py.
Traceback (most recent call last):
  File "train.py", line 184, in <module>
    tf.app.run()
  File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/platform/app.py", line 125, in run
    _sys.exit(main(argv))
  File "/home/user/.local/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py", line 306, in new_func
    return func(*args, **kwargs)
  File "train.py", line 93, in main
    FLAGS.pipeline_config_path)
  File "/media/user/DATA/tensorflow1/models/research/object_detection/utils/config_util.py", line 94, in get_configs_from_pipeline_file
    text_format.Merge(proto_str, pipeline_config)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 536, in Merge
    descriptor_pool=descriptor_pool)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 590, in MergeLines
    return parser.MergeLines(lines, message)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 623, in MergeLines
    self._ParseOrMerge(lines, message)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 638, in _ParseOrMerge
    self._MergeField(tokenizer, message)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 763, in _MergeField
    merger(tokenizer, message, field)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 837, in _MergeMessageField
    self._MergeField(tokenizer, sub_message)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 763, in _MergeField
    merger(tokenizer, message, field)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 837, in _MergeMessageField
    self._MergeField(tokenizer, sub_message)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 763, in _MergeField
    merger(tokenizer, message, field)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 837, in _MergeMessageField
    self._MergeField(tokenizer, sub_message)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 763, in _MergeField
    merger(tokenizer, message, field)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 837, in _MergeMessageField
    self._MergeField(tokenizer, sub_message)
  File "/home/user/.local/lib/python3.6/site-packages/google/protobuf/text_format.py", line 730, in _MergeField
    (message_descriptor.full_name, name))
google.protobuf.text_format.ParseError: 163:9 : Message type "object_detection.protos.LearningRate" has no field named "cosine_decay_google".

[Burgomehl]

Maybe it helps if you do this steps before training: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/exporting_models.md

[anonym24]

@Burgomehl what are you talking about?
this step should be done after training (when you have checkpoints)
all of this described in readme

[Burgomehl]

Yes it is but ... the models are trained on other versions. You are able to bypass some bugs if you export the model before you train ... And I just saied it could help. I don't know if it truely helps.
You are welcome.

[anonym24]

@Burgomehl it generates .pb files and some other files - you do it after you retrained some pretrained model with your classes

[Burgomehl]

Take the help or don't.

"Our frozen inference graphs are generated using the v1.8.0 release version of Tensorflow and we do not guarantee that these will work with other versions; this being said, each frozen inference graph can be regenerated using your current version of Tensorflow by re-running the exporter, pointing it at the model directory as well as the corresponding config file in samples/configs. "

https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

[anonym24]

@Burgomehl
it doesn't create different checkpoints
they are the same after conversation
https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/exporting_models.md

from your link:

what is the point using the same copied checkpoints?

[anonym24]

seems the errors are related to memory issues, though I have 6GB GTX 1060GT..

seems it's better to resize all images to 300x300 before training
and play with batch_size value (default 24)

[anonym24]

yes example faster_rcnn_inception_v2_pets had 1 value for batch_size
others had 24

[anonym24]

though it still fails but after some steps now:

tensorflow/models#2034 (comment)

[anonym24]

training SSD Model works fine with model_train.py on Windows:

#184
tensorflow/models#5719 (comment)