// checkpoint. OR
// there exists a checkpoint inside our log directory. For example, if we are
// using --continue_learning and want to overwrite the saved checkpoint
restoring = load_model_dir or tf.train.latest_checkpoint(checkpoint_dir)
if restoring:
scaffold = TransferScaffold(
local_init_op=tf.group(tf.local_variables_initializer(), init_data_layer)
)
After Change
// directory containing the checkpoint we want to load partial or all weights
// from.. Useful for transer learning or if we do not want to overwrite our
// checkpoint.
restoring = load_model_dir and not tf.train.latest_checkpoint(checkpoint_dir)
if restoring:
scaffold = TransferScaffold(
local_init_op=tf.group(tf.local_variables_initializer(), init_data_layer)
)
