TF2.3 Converting SSD Mobilenet v2 to tflite (tflite file size about 0.5kbytes)

[mpa74]

Prerequisites

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

• I am using the latest TensorFlow Model Garden release and TensorFlow 2.
• I am reporting the issue to the correct repository. (Model Garden official or research directory)
• 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/

2. Describe the bug

After successful training SSD mobilenet v2 fpn-320 model on my own data (inference testing on last checkpoint is OK), i used export_tflite_graph_tf2.py and Python API to get TFlite model. But as result tflite generated with 544 byte size and no inference possibility

3. Steps to reproduce

• use labeled dataset for 1 class (more classes not tested)

• download mobilenetv2 fpn 320 config from link

• in config change: num_classes = 1, fine_tune_checkpoint: "/content/models/research/deploy/ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8/checkpoint/ckpt-0", label_map_path and input path for train_input_reader and eval_input_reader

• train model: !python /content/models/research/object_detection/model_main_tf2.py \ --pipeline_config_path={pipeline_file} \ --model_dir={model_dir} \ --alsologtostderr \ --num_train_steps={12000} \ --sample_1_of_n_eval_examples=1 \ --num_eval_steps={500}

• convert to savedmodel using last ckpt as described in link
  python /content/models/research/object_detection/export_tflite_graph_tf2.py \ --pipeline_config_path /content/models/research/deploy/pipeline_file.config \ --trained_checkpoint_dir /content/training \ --output_directory tflite

• get basic TFLite model
  converter = tf.lite.TFLiteConverter.from_saved_model('tflite/saved_model/) converter.optimizations = [tf.lite.Optimize.DEFAULT] tflite_model = converter.convert() with open('model_optimize2.tflite', 'wb') as f: f.write(tflite_model)

• TFLite generated without any error, but have size 544 bytes
  

4. Expected behavior

Much larger tflite file with inference possibility

5. Additional context

I have many warnings while converting checkpoint to saved graph

6. System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Google Colab
• TensorFlow installed from (source or binary): 2.3
• TensorFlow version (use command below): 2.3
• Python version: 3.7
• Bazel version (if compiling from source): not used
• GCC/Compiler version (if compiling from source): not used
• CUDA/cuDNN version: 10.1
• GPU model and memory: Tesla T4

[saikumarchalla]

@mpa74 Could you please go through this link and let us know if it helps.Thanks!

[mpa74]

@mpa74 Could you please go through this link and let us know if it helps.Thanks!

Thanks for response. I try to get tflite by way in this thread (only converting, without quantizing) and got same unusable tflite (500 bytes size)
Code i used:
tflite_convert --saved_model_dir=train_model/export2/saved_model --output_file=detect_local.tflite

[srjoglekar246]

@mpa74 Are you using the latest TF nightly for TFLite conversion? These features were added recently, so I don't think they have made it to the stable version yet.

[mpa74]

@mpa74 Are you using the latest TF nightly for TFLite conversion? These features were added recently, so I don't think they have made it to the stable version yet.

Tested on nightly versions of tensorflow and tf object detection. Everything is OK. Thanks!

I do:

• created new vurtualenv
  - pip install tf-nighly (tf_nightly-2.4.0.dev20201020-cp37-cp37m-win_amd64 installed), tf-slim, scipy
• git clone https://github.com/tensorflow/models.git
• navigate to models\object_detection
• Get-ChildItem object_detection/protos/*.proto | foreach {./protoc "object_detection/protos/$($_.Name)" --python_out=.} (from PowerShell)
• python setup.py build
• python setup.py install
• pip install tf-models-nightly (because i have error "No module named 'official")
• test installation: python object_detection/builders/model_builder_tf2_test.py - everithing is OK.
• next i retrained model, get SavedModel and convert to tflite using command above.

[srjoglekar246]

Awesome. Thanks for confirming!

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No