rubenvillegas / iclr2017mcnet Goto Github PK

Hello, I think the pre-trained model already been removed, every time when I run "./models/paper_models/download.sh" it always gives me a 403 forbidden error.

I am getting the following error when trying to evaluate a model that I trained:

Traceback (most recent call last):
  File "/home/szetor/Documents/video_prediction/iclr2017mcnet/src/test_kth.py", line 182, in <module>
    main(**vars(args))
  File "/home/szetor/Documents/video_prediction/iclr2017mcnet/src/test_kth.py", line 51, in main
    save_path = quant_dir+"results_model="+best_model+".npz"
TypeError: cannot concatenate 'str' and 'NoneType' objects

I assume at some point before that line, you're supposed to set best_model to "MCNET.model-<iter_num>", but that never happens. I suspect the same bug is in test_ucf101.py.

Hi Ruben,

I would like to recommend adding a few dependencies that were not mentioned in the main README file:

# Needed to run test code
pip install opencv-python
# Needed to download Sports1M
pip install pytube
# Needed to extract UCF-100
sudo apt-get install unrar

The command to extract the UCF-100 .rar file may be helpful as well:

unrar x UCF101.rar

Finally, instructions on reading the .npz files would be useful:

import numpy as np
a = np.load('results_model=MCNET.model-102502.npz')
print(a['psnr'])
print(a['ssim'])

Hi,
The paper says that "Sub-clips for running, jogging and walking were manually trimmed to ensure humans are always present in the frames."
Does this mean that the frames having partially visible human (mostly the beginning and the end frames) have been trimmed to?

Thanks

It is a wonderful work! I notice another paper "Learning to Generate Long-term Future via Hierarchical Prediction" is your job, too. But you don't release code for that paper. Will you release in the future? If you will, can you release it as soon as possible? Or can you just tell me how you process the Penn Action dataset in that paper?

Hi,
I'm trying to load the pretrained model (on S1M dataset) you've provided and train it further on another dataset (PENN) instead of starting from scratch. But when creating MCNET model, if I pass is_train=True, I get an error that the checkpoint doesn't have all the variables.
NotFoundError (see above for traceback): Restoring from checkpoint failed. This is most likely due to a Variable name or other graph key that is missing from the checkpoint. Please ensure that you have not altered the graph expected based on the checkpoint.

Can you kindly provide a pretrained model which can be loaded and trained further? Or can I make some changes to the code to achieve that?

When I ran this code, I encountered the following problem: RuntimeError: The ffmpeg plugin does not work on Python 2.x,but when I input: python then: import ffmpeg,there is no problem,I can't figure out the reason, can someone help me?thx~

Hi Ruben,

Thank you for the MCNet code. I want to use your MCNet for video frame extrapolation task for my custom dataset. It would be great if you suggest me some approach in doing it, or share me some generalized code to work for my dataset.

Regards
Sharath

when i run test_kth.py, it is show AttributeError: 'module' object has no attribute 'compute_ssim'. it seems

something is wrong with it

@rubenvillegas , Is it possible to train and test the mcnet for different values of K and T. I have trained and tested for RGB dataset for the given parameters of K=4 and T=1, but when I change the K and T parameters, the model is not building, do you suggest us to do any configurations to the model.

requirement add python2

Thank you very much.

Hi, according to the paper, the input are the feature maps till the 3rd pooling layer from VGG-16. I understand the motion_enc serves this purpose. I am wondering the pre-trained model includes the weights from pretrained VGG-16 on imagenet or you train it from scratch? Thanks for your time