-
Notifications
You must be signed in to change notification settings - Fork 65
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Pretrained Kth Model checkpoint tensor shapes does not match #22
Comments
Getting a very similar error, would appreciate help! EDIT: We had a dimensions issue before this, with the batch size set to 8, the program gave an error that it should completely divide the dataset with size 819, so we changed the batch size to 9. |
Hi @alexlee-gk, I am having a similar problem as well with the |
The errors from @brandonhuo and @ShreyasKolpe are the same. The problem is that the pre-trained model was trained with grayscale images that had 3 channels (same values tiled for the RGB channels) whereas now the grayscale images have 1 channel. I have new KTH models that were trained with the newer dataset format (and also perform better), but I haven't uploaded them yet. I'll do that soon. In the meantime, if you want to use the old pre-trained models, you can use grayscale images with 3 channels by commenting out this line: https://github.com/alexlee-gk/video_prediction/blob/master/video_prediction/datasets/kth_dataset.py#L118 The issue pointed out by @itstaby is different and that's just a limitation of the current implementation: the size of the evaluation dataset should be divisible by the batch size. |
@alexlee-gk Thank you very much for the reply! After commenting out L118 in kth_dataset.py script, I still have the following error:
What could be this error comes from? Thanks again for the help! |
Hi @alexlee-gk I am having a similar problem as well with the kth dataset and the ours_savp model. when I try to train or evaluate, it always told me "ValueError: as_list() is not defined on an unknown TensorShape. input_shape = inputs.get_shape().as_list()". how can I solve the error?What could be this error comes from? Thanks for the help! |
@brandonhuo @Chuckie-He You need to run the download_and_preprocess_dataset.sh script after commenting out the Line#118 in kth_dataset.py file. And then run the code. It will work. |
I also met the problem of "ValueError: as_list() is not defined on an unknown TensorShape. input_shape = inputs.get_shape().as_list()", the method which commenting out the Line#118 in kth_dataset.py file seems have no effect to solve this problem. |
me too,whether commenting out the Line#118 or not,it doesn’t work。I am very puzzled。 |
Yes, I met this problems for both train and predictions steps. |
hello @wangwen39 @Chuckie-He |
Hi, I switched to tf 1.10 and at least testing is working |
Hi alex @alexlee-gk ,
When I try to load the pretrained KTH model (savp, gan and all of them) for testing, there is an error:
It seems the checkpoint cannot be restored for the testing. I would very appreciate your help!
The entire error message is below:
error.txt
The text was updated successfully, but these errors were encountered: