Deep Multi Scale Video Prediction Beyond Mean Square Error

Authors: Michael Mathieu, camille couprie, Yann Lecun

ICLR 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We compare our predictions to different published results based on recurrent neural networks on the UCF101 dataset.
Researcher Affiliation Collaboration 1New York University 2Facebook Artificial Intelligence Research
Pseudocode Yes Algorithm 1: Training adversarial networks for next frame generation
Open Source Code No The paper mentions a link 'http://cs.nyu.edu/ mathieu/iclr2016.html' for 'our results', but not for source code.
Open Datasets Yes We use the Sports1m for the training, because most of UCF101 frames only have a very small portion of the image actually moving... from the Sports1m dataset of Karpathy et al. (2014) and UCF101 (Soomro et al., 2012).
Dataset Splits No The paper mentions training on Sports1m and testing on UCF101, but does not specify a separate validation dataset or split. It only refers to 'test' evaluations.
Hardware Specification No The paper mentions taking advantage of 'GPU hardware capabilities' but does not specify any particular GPU models, CPU types, or other detailed hardware specifications used for the experiments.
Software Dependencies No The paper mentions using 'Matlab code' for Epic Flow computation, but does not provide specific version numbers for its own implemented software dependencies (e.g., deep learning frameworks like PyTorch/TensorFlow, Python version, etc.).
Experiment Setup Yes The learning rate ρG starts at 0.04 and is reduced over time to 0.005. The minibatch size is set to 4, or 8 in the case of the adversarial training, to take advantage of GPU hardware capabilities. We train the network on small patches... The network is trained by setting the learning rate ρD to 0.02.