[Solved] Problems with TensorFlow 1.0 and Windows

[drfknoble]

Hi there,

First, I wanted to say thanks for sharing! I'm working through the code to help with my own segmentation project and having something to work from is a big help.

Second, I came across a few issues (minor really) that I've figured out and wanted to share:

• TensorFlow 1.0 replaced tf.pack() with tf.stack().
• In TensorFlow 1.0, variables should be initialised using tf.global_variables_initializer()
• In Windows, the os.path.splittext() should use "\ \", rather than '/'. Otherwise, the program can't find any files to pickle (and the MITSceneParsing.pickle file is empty), which in turn means 0 records are found and the feed dict instruction doesn't work.

Like I said, pretty minor stuff, but I wanted to post in case anyone else had any issues.

Best regards,

Frazer

P.S. If you get an out of memory error, it's likely because you're trying to work with 20,000 images, which might be a bit too much. I deleted some of the training images and it worked.

[jiao0805]

Hi

Can this project be successfully run on windows ?
My tensorflow is 1.0 gpu version on windows.

[drfknoble]

Hi Jiao0805,

Absolutely, I've trained this neural network on a Windows 8 and a Windows 10 Machine, using both TensorFlow CPU and GPU versions. I'd recommend you use Anaconda to manage your Python environments and use PIP to install the version you're after.

Here's a sample output from my application, which detects lanes on a road. For reference, I used the KITTI data set (see http://www.cvlibs.net/datasets/kitti/).

For my Windows 8 machine, I had to run TensorFlow on my CPU, as my GPU doesn't have enough memory (you'll need > 4 GB for this repo' to run). Last, but not least, give it at least 30, 000 training steps to get reasonable outputs; 10, 000, as a minimum, should start to give tentative results.

Best regards,

Frazer

[jiao0805]

Hi FKNoble,
Thanks a lot for your reply!
I use tensorflow 1.0 gpu on windows 7 , and I saw the issues you shared above.
I did all of them but I still got one error as follows:

Traceback (most recent call last):
File "FCN.py", line 221, in
tf.app.run()
File "C:\Users\SEELE\AppData\Local\Programs\Python\Python35\lib\site-packages
tensorflow\python\platform\app.py", line 44, in run
_sys.exit(main(_sys.argv[:1] + flags_passthrough))
File "FCN.py", line 152, in main
loss = tf.reduce_mean((tf.nn.sparse_softmax_cross_entropy_with_logits(logits
,tf.squeeze(annotation, squeeze_dims=[3]),name="entropy")))
File "C:\Users\SEELE\AppData\Local\Programs\Python\Python35\lib\site-packages
tensorflow\python\ops\nn_ops.py", line 1684, in sparse_softmax_cross_entropy_wit
h_logits
labels, logits)
File "C:\Users\SEELE\AppData\Local\Programs\Python\Python35\lib\site-packages
tensorflow\python\ops\nn_ops.py", line 1533, in _ensure_xent_args
"named arguments (labels=..., logits=..., ...)" % name)
ValueError: Only call sparse_softmax_cross_entropy_with_logits with named argu
ments (labels=..., logits=..., ...)

It seems that there is something wrong with my cost function.
Did you meet this error on win8?
Could you please kindly tell me how can I solve it?

[drfknoble]

I had something similar and what I think it means is that the cross entropy function needs the input parameters to be explicitly labeled, i.e. labels=labels, logits=logits, etc.

Here's the function that I have, which works:

loss = tf.reduce_mean((tf.nn.sparse_softmax_cross_entropy_with_logits(
logits=logits,
labels=tf.squeeze(annotation, squeeze_dims=[3]),
name="entropy")))

Let me know if it fixes your problem.

Best regards,

Frazer

[jiao0805]

Hi FKNoble,
I fixed my problem and I found that the cross entropy function actually needs four parameters.

Please have a look on this:

def _ensure_xent_args(name, sentinel, labels, logits):
#Make sure that all arguments were passed as named arguments.
if sentinel is not None:
raise ValueError("Only call %s with "
"named arguments (labels=..., logits=..., ...)" % name)

I found them in the declaration of tf.nn.sparse_softmax_cross_entropy_with_logits()
I tried your code and my problem fixed.
I also tried this:

None,
tf.squeeze(annotation, squeeze_dims=[3]),
logits,
"entropy")))
And it also worked.

And as for the os.path.splittext()
To split URL, use '/' , to split file path, use'\'
otherwise program can not found the file.
see here:

SceneParsing_folder = os.path.splitext(DATA_URL.split("/")[-1])[0]
filename = os.path.splitext(f.split("\\")[-1])[0] 

And I finally run into a error as follows:

Setting up Saver...
Step: 0, Train_loss:316.906
2017-03-14 12:25:00.698600 ---> Validation_loss: 424.702
E c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\stream_executor\cuda\cuda_event.cc:49] Error polling for event status: failed to query event: CUDA_ERROR_UNKNOWN
F c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\common_runtime\gpu\gpu_event_mgr.cc:198] Unexpected Event status: 1
E c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\stream_executor\cuda\cuda_driver.cc:1034] failed to alloc 536870912 bytes on host: CUDA_ERROR_UNKNOWN
W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\common_runtime\gpu\pool_allocator.h:195] could not allocate pinned host memory of size: 536870912
...
...
...
E c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\stream_executor\cuda\cuda_driver.cc:1034] failed to alloc 536870912 bytes on host: CUDA_ERROR_UNKNOWN
W c:\tf_jenkins\home\workspace\release-win\device\gpu\os\windows\tensorflow\core\common_runtime\gpu\pool_allocator.h:195] could not allocate pinned host memory of size: 536870912

Is that mean that I do not have enough memory?
My graphics card has 6GB memory.

Looking for your reply!

Best regards,
Jiao

[shekkizh]

Thanks @FKNoble for helping out.

@jiao0805 Your last error does seem like an issue with insufficient gpu memory. try reducing batch size or image size. Also it would be helpful if you post different issues in a separate thread - it will be useful for people in the future to make note of. Thanks.

[BowieHsu]

@FKNoble thanks for your sharing, but after i modified my code as you said, when i try to training the network, i still got loss:Nan results, my running environment is windows 7 + TITAN X + python3.5 + tensorflow 1.01, training set is ADEChallengeData2016 with num_of_classes = 151

[chankeh]

I used the version 0.12rc, but it occured like this：

Traceback (most recent call last):
Found pickle file!
File "E:/Python/FCN/FCN.py", line 224, in
tf.app.run()
File "D:\Program Files\Anaconda3\lib\site-packages\tensorflow\python\platform\app.py", line 43, in run
sys.exit(main(sys.argv[:1] + flags_passthrough))
File "E:/Python/FCN/FCN.py", line 168, in main
train_records, valid_records = scene_parsing.read_dataset(FLAGS.data_dir)
File "E:\Python\FCN\read_MITSceneParsingData.py", line 30, in read_dataset
training_records = result['training']
TypeError: 'NoneType' object is not subscriptable

how can i fixed this problem？
thanks.

[CLegendBuptSun]

Sorry to bother you.
When I test the random batch(visualization),I get the answers of nearly black prediction mask，I read your code top to the bottom，but didnot get the reason。
Then，I changed the batchsize=16 and learning rate=1e-4， MAX_ITERATION =50000，finally，I got
something but nearly close to black to...
I want to know WHY.
thank you for your reply.

[drenego]

I think you can try with batchsize=1 and learning rate=1e-5 max_iteration>=100000, then you will get an acceptable model, at least i did this.

[zhshining]

Hi FKNoble
Sorry to trouble you .
I ran FCN.py in my windows by cpu Tensorflow, but there were some questions :
Traceback (most recent call last):
File "D:/DL/data/FCN/FCN.tensorflow-master/FCN.py", line 223, in
tf.app.run()
File "D:\DL\Tensorflow\anconda\lib\site-packages\tensorflow\python\platform\app.py", line 48, in run
_sys.exit(main(_sys.argv[:1] + flags_passthrough))
File "D:/DL/data/FCN/FCN.tensorflow-master/FCN.py", line 194, in main
sess.run(train_op, feed_dict=feed_dict)
File "D:\DL\Tensorflow\anconda\lib\site-packages\tensorflow\python\client\session.py", line 889, in run
run_metadata_ptr)
File "D:\DL\Tensorflow\anconda\lib\site-packages\tensorflow\python\client\session.py", line 1096, in _run
% (np_val.shape, subfeed_t.name, str(subfeed_t.get_shape())))
ValueError: Cannot feed value of shape (0,) for Tensor 'input_image:0', which has shape '(?, 224, 224, 3)'

I researched less information about it in Web. I don't know weather codes' locations are wrong or not.
How can i fixed this problems.
Thanks a lot.