Quadratic Video Interpolation

Authors: Xiangyu Xu, Li Siyao, Wenxiu Sun, Qian Yin, Ming-Hsuan Yang

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we first provide implementation details of the proposed model, including training data, network structure, and hyper-parameters. We then present evaluation results of our algorithm with comparisons to the state-of-the-art methods on video datasets. The source code, data, and the trained models are available at: https://sites.google.com/view/xiangyuxu/qvi_nips19.
Researcher Affiliation Collaboration Xiangyu Xu Carnegie Mellon University xuxiangyu2014@gmail.com Li Siyao Sense Time Research lisiyao1@sensetime.com Wenxiu Sun Sense Time Research sunwenxiu@sensetime.com Qian Yin Beijing Normal University yinqian@bnu.edu.cn Ming-Hsuan Yang University of California, Merced Google mhyang@ucmerced.edu
Pseudocode No The paper describes the algorithm steps in text and diagrams but does not include any formal 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes The source code, data, and the trained models are available at: https://sites.google.com/view/xiangyuxu/qvi_nips19.
Open Datasets Yes We evaluate our model with the state-of-the-art video interpolation approaches, including the phase-based method (Phase) [17], separable adaptive convolution (Sep Conv) [21], deep voxel flow (DVF) [14], and Super Slo Mo [9]. ... high frame rate video datasets such as GOPRO [18] and Adobe240 [30]. We also conduct experiments on the UCF101 [29] and DAVIS [23] datasets...
Dataset Splits No The paper describes the datasets used for training and testing, and how data is prepared (e.g., 'randomly crop 352 352 patches for training'), but it does not specify explicit training/validation/test splits with percentages or counts for a distinct validation set.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions using specific optimizers and network architectures (e.g., 'Adam optimizer', 'PWC-Net', 'U-Net') but does not specify version numbers for any software dependencies.
Experiment Setup Yes We initialize the learning rate as 10 4 and further decrease it by a factor of 0.1 at the end of the 100th and 150th epochs. The trade-off parameter λ of the loss function (7) is set to be 0.005. k in the activation function of δ is set to be 10. In the flow reversal layer, we set the Gaussian standard deviation σ = 1. We first train the proposed network with the flow estimation module fixed for 200 epochs, and then finetune the whole system for another 40 epochs.