search

vonclites / squeezenet

Tensorflow implementation of SqueezeNet.
MIT License
129 stars 63 forks source link

Hyperparams arguments missing #15

Closed HaFred closed 4 years ago

HaFred commented 4 years ago 
Squeezenet Training Program: error: the following arguments are required: --model_dir, --train_tfrecord_filepaths, --validation_tfrecord_filepaths, --network, --num_classes, --num_gpus, --batch_size, --num_input_threads, --shuffle_buffer

Hi, does anybody know how to set the arguments? I did not find relating files in the repo, for example the tfrecord_filepaths...

vonclites commented 4 years ago

These are command line arguments. So on the command line you would run python train_squeezenet.py --model_dir=<path_to_model_dir> etc. All the parameters are specified in https://github.com/vonclites/squeezenet/blob/master/arg_parsing.py However, this was implemented using a very old version of Tensorflow... the API it uses might be deprecated by now. You may be better off using it as a guideline for a reimplementation in Tensorflow 2.

vonclites commented 4 years ago

Also, the tfrecords aren't provided in this repository. You would have to create them. See https://github.com/vonclites/squeezenet/blob/f13555462e835409480829f8fdacac96835161e5/inputs.py#L169 for the fields the tfrecord examples are expected to have (an 'image' represented as a byte string and an int64 label).