Video future frames prediction based on Transformers. Published on ICPR2022, https://ieeexplore.ieee.org/abstract/document/9956707
Published on Image and Vision Computing. https://arxiv.org/pdf/2212.06026.pdf
The overall framework for video prediction.
Fully autoregressive (left) and non-autoregressive VPTR (right).
Download the checkpoints from here: https://polymtlca0-my.sharepoint.com/:f:/g/personal/xi_ye_polymtlus_ca/EuxjSddJ7wNIsiSTOfB-u7AB7qQhP5H0iX2a5mbaowiSZw?e=taglte
See Test_AutoEncoder.ipynb and Test_VPTR.ipynb for the detatiled test functions.
Train the autoencoder firstly, save the ckpt, load it for stage 2
train_FAR.py: Fully autoregressive model
train_FAR_mp.py: multiple gpu training (single machine)
train_NAR.py: Non-autoregressive model
train_NAR_mp.py: multiple gpu training (single machine)
/MovingMNIST
moving-mnist-train.npz
moving-mnist-test.npz
moving-mnist-val.npz
/KTH
boxing/
person01_boxing_d1/
image_0001.png
image_0002.png
...
person01_boxing_d2/
image_0001.png
image_0002.png
...
handclapping/
...
handwaving/
...
jogging_no_empty/
...
running_no_empty/
...
walking_no_empty/
...
/BAIR
test/
example_0/
0000.png
0001.png
...
example_1/
0000.png
0001.png
...
example_...
train/
example_0/
0000.png
0001.png
...
example_...
Please cite the paper if you find our work is helpful.
@inproceedings{ye2022vptr,
title={VPTR: Efficient Transformers for Video Prediction},
author={Ye, Xi and Bilodeau, Guillaume-Alexandre},
booktitle={2022 26th International Conference on Pattern Recognition (ICPR)},
pages={3492--3499},
year={2022},
organization={IEEE}
}
@article{ye2022video,
title={Video prediction by efficient transformers},
author={Ye, Xi and Bilodeau, Guillaume-Alexandre},
journal={Image and Vision Computing},
pages={104612},
year={2022},
publisher={Elsevier}
}
Recently, we found a mistake in our ICPR paper. For the BAIR experiments, the previous papers predict 28 future frames instead of 10. Specifically, the results in "TABLE II: Results on BAIR" are for 10 future frames instead of 28. The results for 28 predicted frames are updated here, see the following correct table.
We apologize for the mistake, the correction does not affect our conclusions.