using scripts/retrain.py how to proceed?

[monajalal]

I have a naive question. I followed the Google Collab tutorial for transfer learning on images of 5 categories of flowers. However, how can I do the same for the images of hands from egohands? What are the changes I should perform?

mona@Mona:~/code/handpose/handtracking$ IMAGE_SIZE=224
mona@Mona:~/code/handpose/handtracking$ ARCHITECTURE="mobilenet_0.50_${IMAGE_SIZE}"
mona@Mona:~/code/handpose/handtracking$ python -m scripts.retrain   --bottleneck_dir=tf_files/bottlenecks   --how_many_training_steps=500   --model_dir=tf_files/models/   --summaries_dir=tf_files/training_summaries/"${ARCHITECTURE}"   --output_graph=tf_files/retrained_graph.pb   --output_labels=tf_files/retrained_labels.txt   --architecture="${ARCHITECTURE}"   --image_dir=images/train
/home/mona/anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
>> Downloading mobilenet_v1_0.50_224_frozen.tgz 100.1%
--- Logging error ---
Traceback (most recent call last):
  File "/home/mona/anaconda3/lib/python3.6/logging/__init__.py", line 992, in emit
    msg = self.format(record)
  File "/home/mona/anaconda3/lib/python3.6/logging/__init__.py", line 838, in format
    return fmt.format(record)
  File "/home/mona/anaconda3/lib/python3.6/logging/__init__.py", line 575, in format
    record.message = record.getMessage()
  File "/home/mona/anaconda3/lib/python3.6/logging/__init__.py", line 338, in getMessage
    msg = msg % self.args
TypeError: not all arguments converted during string formatting
Call stack:
  File "/home/mona/anaconda3/lib/python3.6/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "/home/mona/anaconda3/lib/python3.6/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/home/mona/code/handpose/handtracking/scripts/retrain.py", line 1326, in <module>
    tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)
  File "/home/mona/anaconda3/lib/python3.6/site-packages/tensorflow/python/platform/app.py", line 126, in run
    _sys.exit(main(argv))
  File "/home/mona/code/handpose/handtracking/scripts/retrain.py", line 982, in main
    maybe_download_and_extract(model_info['data_url'])
  File "/home/mona/code/handpose/handtracking/scripts/retrain.py", line 339, in maybe_download_and_extract
    'bytes.')
  File "/home/mona/anaconda3/lib/python3.6/site-packages/tensorflow/python/platform/tf_logging.py", line 116, in info
    _get_logger().info(msg, *args, **kwargs)
Message: 'Successfully downloaded'
Arguments: ('mobilenet_v1_0.50_224_frozen.tgz', 6308169, 'bytes.')
ERROR:tensorflow:No valid folders of images found at images/train

It does seem I am not the only one who is needing help with this based on other git issues. Some guidance is really appreciated.

[victordibia]

Hi Mona,

Transfer learning for image classification and object detection tasks differ. It seems the flower tutorial you refer to is a classic image classification task, while the hand tracking is an object identification task. I found the stanford cs231n course helpful in building an understanding of these tasks.

Now, regarding training, the work here heavily depends on the Tensorflow Object Detection Api and assumes the reader is familiar with the api. I would suggest running the api code and understanding it.
I also provide provide reference to tutorials I used/followed to do my own "transfer learning". Please see issue #18

Goodluck!

-V.