While upgrade TensorFlow to 1.5, then training own data-set throwing the following error

[Bahramudin]

System information

• What is the top-level directory of the model you are using: object detection
• Have I written custom code (as opposed to using a stock example script provided in TensorFlow):
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 16.04
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): v1.5.0-0-g37aa430d84 1.5.0
• Bazel version (if compiling from source):
• CUDA/cuDNN version: using CPU
• GPU model and memory: using CPU
• Exact command to reproduce:

Describe the problem

I want to train my own dataset using TensorFlow object detection. I have already done everything step by step (collecting data, mark objects, create TF records) it was done without any problem, and when I was using TensorFLow 1.4 to train the dataset by using ssd_mobilenet_v1_coco_2017_11_17, everything goes well with out any problem, but when I upgrade it to 1.5, then I am getting below error:

WARNING:tensorflow:From /home/adil/workspace/tensorflow/models-working/research/object_detection/trainer.py:228: create_global_step (from tensorflow.contrib.framework.python.ops.variables) is deprecated and will be removed in a future version.
Instructions for updating:
Please switch to tf.train.create_global_step
Traceback (most recent call last):
File "train.py", line 167, in
tf.app.run()
File "/home/adil/.local/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 124, in run
_sys.exit(main(argv))
File "train.py", line 163, in main
worker_job_name, is_chief, FLAGS.train_dir)
File "/home/adil/workspace/tensorflow/models/research/object_detection/trainer.py", line 235, in train
train_config.prefetch_queue_capacity, data_augmentation_options)
File "/home/adil/workspace/tensorflow/models/research/object_detection/trainer.py", line 59, in create_input_queue
tensor_dict = create_tensor_dict_fn()
File "train.py", line 120, in get_next
dataset_builder.build(config)).get_next()
File "/home/adil/workspace/tensorflow/models/research/object_detection/builders/dataset_builder.py", line 138, in build
label_map_proto_file=label_map_proto_file)
File "/home/adil/workspace/tensorflow/models-working/research/object_detection/data_decoders/tf_example_decoder.py", line 110, in init
dct_method=dct_method),
TypeError: init() got an unexpected keyword argument 'dct_method'

Note: I have already updated (reinstalled) everything.

[Paspartout]

I had the same problem since yesterday.

It looks like the pull request merged in be9b802
caused some problems. After checking out the version before that commit everything worked fine.

[Bahramudin]

@Paspartout Hmmm, today when I installed everything from the latest (TensorFlow and Models) everything goes well, but while starting to train, got this problem, but I have tried to solve it by my self or by searching on the internet, but none of them has solved the problem. Here is a stack overflow question also issuing the same problem.

[RonakDedhiya]

Facing out same problem

[Lisandro79]

Same problem here.

[ghost]

This error is due to the incompatibility of file tf_example_decoder.py with the Tensorflow installed.

In file models/research/object_detection/data_decoders/tf_example_decoder.py:

line 28:

slim_example_decoder = tf.contrib.slim.tfexample_decoder

line 104:

self.items_to_handlers = {
    fields.InputDataFields.image:
        slim_example_decoder.Image(
            image_key='image/encoded',
            format_key='image/format',
            channels=3,
            dct_method=dct_method),....
            .....
            ....
            ..
            .
            }

slim_example_decoder.Image is tf.contrib.slim.tfexample_decoder.Image

In Tensorflow 1.5 source code tf.contrib.slim.tfexample_decoder.Image doesn't have dct_method argument. Therefore when dct_method argument is passed it throws an error.

A simple hack would be to not pass this argument. So edit tf_example_decoder.py line 110 to remove dct_method=dct_method. I tried this and it worked and didn't had any effect on training.

[bignamehyp]

@nathansilberman can you please take a look? Thanks

[pkulzc]

@harshitladdha is right, this code depends on latest tf.contrib.slim.tfexample_decoder, which does not exist in tf 1.5. A fix will be applied soon.

[nguyeho7]

Another workaround is to install tf-nightly (version 1.7)

[ybsave]

But there is no recent nightly built version for Windows users that has the 'dct_method' version. The last successful windows nightly built is almost one month ago!

[ymodak]

@Bahramudin Can you please use the latest version of TensorFlow and test again? Thanks!

[pkulzc]

Please note that Tesnorflow Object Detection now requires tensorflow v 1.9 +.

Closing this since it's we have updated our code base a number of times.