Anyway to fine-tune?

[hiepph]

I have a small subset (thousand at max) of coordinates of word bounding boxes. I don't want to train from scratch, but rather fine-tune from weights pre-trained by the icdar2015 dataset.

Is it possible, and how?

[argman]

you can load the weights from the ckpt i provide, just replace the imagenet cktp with icdar ckpt

[hao-liu-china]

I replace the pretrained model path to ckpt you provide, but the weights under a given path disappear in the begining of training. Why? Can you give us some specific details? like python........

[hiepph]

@argman I load the weights model.ckpt-49491 as --pretrained_model_path, but the test result is pretty bad (it cannot detect any text boxes). What is the problem here?

[JmmSu]

@mozhuangb Remember to change the path in file"checkpoint"

[YanShuang17]

I want to initialize model with the trained icdar2015 model provided by author, and just set the parameter below is:

tf.app.flags.DEFINE_boolean('restore', True, 'whether to resotre from checkpoint')

and run multigpu_train.py, but error occured like below:

continue training from previous checkpoint
ERROR:tensorflow:Couldn't match files for checkpoint /tmp/east_icdar2015_resnet_v1_50_rbox/model.ckpt-49491
Traceback (most recent call last):
  File "./multigpu_train.py", line 181, in <module>
    tf.app.run()
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/platform/app.py", line 48, in run
    _sys.exit(main(_sys.argv[:1] + flags_passthrough))
  File "./multigpu_train.py", line 142, in main
    saver.restore(sess, ckpt)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/saver.py", line 1548, in restore
    {self.saver_def.filename_tensor_name: save_path})
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 789, in run
    run_metadata_ptr)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 997, in _run
    feed_dict_string, options, run_metadata)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 1132, in _do_run
    target_list, options, run_metadata)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 1152, in _do_call
    raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.InternalError: Unable to get element from the feed as bytes.

anything wrong with me??? or how to fine-tune? @argman @hiepph @mozhuangb @JmmSu @Atry thanks.

[DunaiFuentes]

@hiepph What worked for me was the following: Instead of using the pretrained_model_path, continue training from the given checkpoint with checkpoint_path = "your_path", with restore=True. Additionally, as @JmmSu noted you have to modify the checkpoint file inside the folder so that it points to its new path.

[Vivianwxf]

@hiepph I have the same confusion with you, have you work out? Please tell me how can I fine-tune my own dataset? Thank you very much. Ｉdon't understand their comments.

[YangZeyu95]

@YanShuang17 change the fist line in file 'checkpoint' from "/tmp/east_icdar2015_resnet_v1_50_rbox/model.ckpt-49491" to "model.ckpt-49491" and add parameter --checkpoint_path=<your_checkpoint_path> eg. --checkpoint_path=checkpoint/fine_tone/

[richaagrawa]

Hi,

When I am trying to do restore=True, the error is
File "multigpu_train.py", line 180, in <module> tf.app.run() File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/platform/app.py", line 40, in run _run(main=main, argv=argv, flags_parser=_parse_flags_tolerate_undef) File "/usr/local/lib/python3.6/dist-packages/absl/app.py", line 299, in run _run_main(main, args) File "/usr/local/lib/python3.6/dist-packages/absl/app.py", line 250, in _run_main sys.exit(main(argv)) File "multigpu_train.py", line 141, in main saver.restore(sess, ckpt) File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/training/saver.py", line 1277, in restore raise ValueError("Can't load save_path when it is None.") ValueError: Can't load save_path when it is None.

I have changed the first line of checkpoint file as : model_checkpoint_path: "model.ckpt-49491" and passing --checkpoint_path=/path/pretrained_model_path.ckpt, but still error persists.

Can you please help with this!!!