Error in computing fid during training. ValueError: Cannot iterate over a shape with unknown rank.

[Ayesha-Rafiq]

No description provided.

[Ayesha-Rafiq]

Epoch: [ 0] [ 0/ 305] time: 14.0901, d_loss: 1.75800323, g_loss: 0.26850128
[Sample] d_loss: 1.03988624, g_loss: 1.53502333
samples for incept 2/2 ok
Traceback (most recent call last):
File "main.py", line 161, in
tf.app.run()
File "C:\Users\Waqar\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\platform\app.py", line 48, in run
_sys.exit(main(_sys.argv[:1] + flags_passthrough))
File "main.py", line 144, in main
dcgan.train(FLAGS)
File "D:\downloads\TTUR-master\TTUR-master\DCGAN_FID_batched\model.py", line 381, in train
verbose=self.fid_verbose)
File "D:\downloads\TTUR-master\TTUR-master\DCGAN_FID_batched\fid.py", line 175, in calculate_activation_statistics
act = get_activations(images, sess, batch_size, verbose)
File "D:\downloads\TTUR-master\TTUR-master\DCGAN_FID_batched\fid.py", line 82, in get_activations
inception_layer = _get_inception_layer(sess)
File "D:\downloads\TTUR-master\TTUR-master\DCGAN_FID_batched\fid.py", line 55, in _get_inception_layer
shape = [s.value for s in shape]
File "C:\Users\Waqar\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 497, in iter
raise ValueError("Cannot iterate over a shape with unknown rank.")
ValueError: Cannot iterate over a shape with unknown rank.

[mhex]

Hi, which Tensorflow version do you use?

[Ayesha-Rafiq]

My tensorflow version is 1.4.0

[Ayesha-Rafiq]

mhex???

[HRamses]

Hi!
Could you try to change to a more recent version of tf? we had to adapt the code for newer tf versions - so hopefully this will fix the problem.

[Ayesha-Rafiq]

I have tried tensorflow version 1.6 but same error occur

…

On Friday, August 10, 2018, HRamses ***@***.***> wrote: Hi! Could you try to change to a more recent version of tf? we had to adapt the code for newer tf versions - so hopefully this will fix the problem. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#10 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AleDwVA2k80HxTPkG2NOqzKJgenvJQ62ks5uPXRCgaJpZM4V3lcS> .

[Ayesha-Rafiq]

When i compute fid seperatly. Then it works fine. But when i try to compute fid during training of dcgan then this error occur. please help me.

…

On Friday, August 10, 2018, AYESHA RAFIQUE ***@***.***> wrote: I have tried tensorflow version 1.6 but same error occur On Friday, August 10, 2018, HRamses ***@***.***> wrote: > Hi! > Could you try to change to a more recent version of tf? we had to adapt > the code for newer tf versions - so hopefully this will fix the problem. > > — > You are receiving this because you authored the thread. > Reply to this email directly, view it on GitHub > <#10 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AleDwVA2k80HxTPkG2NOqzKJgenvJQ62ks5uPXRCgaJpZM4V3lcS> > . >

[byeongkeun-kang]

I have the same problem (both this and #8). @HRamses, Could you tell me the version of Tensorflow you have tested?

[HRamses]

I tested it with version 1.8.0. But I tested it only on an separate data set and not while training. I will try to reproduce the error.

[Ayesha-Rafiq]

I have also tried tensorflow version 1.8.0. but same error occur.

…

On Mon, Aug 13, 2018 at 1:44 PM, HRamses ***@***.***> wrote: I testet it with version 1.8.0 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#10 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AleDwa-P9Cy1f401M-UeegV2EynJwZ0nks5uQTyHgaJpZM4V3lcS> .

[mhex]

It looks like the code is not compatible anymore with the recent Python, Numpy? and Tensorflow versions. We need some time to sort this out.

[untom]

If you need a more short-term solution, please note that you can use the FID implementation from tensorflow contrib: https://www.tensorflow.org/api_docs/python/tf/contrib/gan/eval/frechet_classifier_distance

[paruby]

I also got this error using TF version 1.10 and python version 3.6.

I haven't tested this, but a fix that stops things from crashing and seems to give correct numbers is to replace line 54:

if shape._dims != []:

with:

if shape._dims != [] and (shape._dims is not None):