Video Frame Interpolation via Deformable Separable Convolution

Authors: Xianhang Cheng, Zhenzhong Chen10607-10614

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that our method significantly outperforms the other kernel-based interpolation methods and shows strong performance on par or even better than the state-of-the-art algorithms both qualitatively and quantitatively.
Researcher Affiliation Academia Xianhang Cheng,1 Zhenzhong Chen1 1School of Remote Sensing and Information Engineering, Wuhan University, China {xianhang, zzchen}@whu.edu.cn
Pseudocode No The paper describes methods and architecture in text and diagrams but does not contain a formal pseudocode or algorithm block.
Open Source Code No The paper does not provide a statement about releasing source code for the described methodology or a link to a repository.
Open Datasets Yes Our training dataset is Vimeo90K (Xue et al. 2019). ... For UCF101 dataset, we use 379 triplets chosen by Liu et al. (Liu et al. 2017)... The Middleburry benchmark has been widely used for assessing frame interpolation methods...
Dataset Splits No The paper mentions a training dataset (Vimeo90K) and test sets (Vimeo90K test set, UCF101, Middleburry), but does not explicitly define or refer to a separate validation dataset split.
Hardware Specification No The paper does not provide any specific hardware details (like GPU or CPU models) used for running its experiments.
Software Dependencies No The paper mentions the use of Adam optimizer and ReLU activation, but does not specify any software libraries or their version numbers (e.g., PyTorch, TensorFlow version).
Experiment Setup Yes Our training dataset is Vimeo90K (Xue et al. 2019). The triplets in this dataset were randomly flipped horizontally or vertically for data augmentation. The network was trained using Adam optimizer (Kingma and Ba 2014). We first trained our network for 120 epochs using a learning rate schedule of 1e-4, dropping by half every 40 epochs. The training patch size was randomly cropped into 128 128 pixels and the batch size was 16. Inspired by some previous work training their networks with larger patches (Niklaus and Liu 2018; Bao et al. 2018a; 2019), we fine-tuned our network using patches of size 256 256 and the entire frames with learning rates of 6.25e6 and 3.125e-6, respectively.