Efficient and Information-Preserving Future Frame Prediction and Beyond

Authors: Wei Yu, Yichao Lu, Steve Easterbrook, Sanja Fidler

ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our proposed approach achieves state-of-the-art results on Moving MNIST, Traffic4cast and KITTI datasets.
Researcher Affiliation Collaboration 1Department of Computer Science, University of Toronto 2Vector Institute, Canada 3NVIDIA
Pseudocode No The paper describes the model architecture and mathematical equations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statement about making the source code available or include a link to a code repository for their proposed method.
Open Datasets Yes Moving MNIST (Srivastava et al. (2015)), Traffic4cast (IARAI (2019)), KITTI (Geiger et al. (2012)), and Caltech Pedestrian (Dollár et al. (2009)) are used and cited in the paper.
Dataset Splits No The paper mentions training and evaluating on a fixed test set for Moving MNIST (e.g., 'evaluated with a fixed test set containing 5, 000 sequences') and mentions separate sampling for training and testing digits, but it does not explicitly provide details about a distinct validation set or its split ratios for any of the datasets.
Hardware Specification Yes Our Crev Net does not suffer from high memory consumption so that we were able to train our model in a single V100 GPU.
Software Dependencies No The paper mentions using 'Adam optimizer' and 'SSD (Liu et al. (2016)) as detection head', but it does not specify any version numbers for these or other software components or libraries.
Experiment Setup Yes All variants of Crev Net are trained by using the Adam optimizer with a starting learning rate of 5 10 4 to minimize MSE. The training process is stopped after 300, 000 iterations with the batch size of 16...