Unsupervised Deep Video Denoising with Untrained Network

Authors: Huan Zheng, Tongyao Pang, Hui Ji

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experiments on video denoising demonstrate that the proposed unsupervised method outperforms existing unsupervised methods and remains competitive against recent supervised deep learning methods.
Researcher Affiliation Academia Department of Mathematics at National University of Singapore, Singapore. huan zheng@u.nus.edu, matpt@nus.edu.sg, matjh@nus.edu.sg
Pseudocode No No structured pseudocode or algorithm blocks were found.
Open Source Code Yes The code is available at https://github.com/huanzheng551803/VER2R.
Open Datasets Yes We use two datasets, DAVIS(Khoreva, Rohrbach, and Schiele 2018) and Set8(Tassano, Delon, and Veit 2020). The experiment uses the real raw video dataset (Yue et al. 2020b)
Dataset Splits No The paper mentions 'DAVIS has a training set and a test set' and 'The dataset is divided into a training set and a test set, with the first six video sequences forming the training set and the remaining five forming the test set.' However, it does not explicitly provide validation set splits or specific percentages/counts for all splits across all experiments for full reproducibility.
Hardware Specification No The paper mentions 'All the deep learning methods are conducted using the same computing infrastructure as ours for time counting' but does not provide specific hardware details such as GPU/CPU models, memory, or cloud instance types.
Software Dependencies No The paper states 'Our method is implemented using Py Torch.' but does not provide specific version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes In the first stage of training, the iteration number does not exceed 1500 or 30N (the number of frames). In the second stage, we train the whole network for 50 epochs to denoise each frame sequentially.