Video Decomposition Prior: Editing Videos Layer by Layer

Authors: Gaurav Shrivastava, Ser-Nam Lim, Abhinav Shrivastava

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach on standard video datasets like DAVIS, REVIDE, & SDSD and show qualitative results on a diverse array of internet videos.
Researcher Affiliation Academia Gaurav Shrivastava University of Maryland, College Park gauravsh@umd.edu Ser-Nam Lim University of Central Florida sernam@ucf.edu Abhinav Shrivastava University of Maryland, College Park abhinav@cs.umd.edu
Pseudocode No The paper describes network architectures and optimization steps in text and tables but does not provide a formal pseudocode block or algorithm.
Open Source Code No The paper does not provide an explicit statement about releasing code for the described methodology or a link to a code repository.
Open Datasets Yes We evaluate our approach on standard video datasets like DAVIS, REVIDE, & SDSD and show qualitative results on a diverse array of internet videos.
Dataset Splits No The paper describes an inference-time optimization framework that optimizes directly on the test sequence itself, rather than training on a dataset with traditional train/validation splits.
Hardware Specification Yes To optimize our model, we use a single Nvidia A6000 GPU with 48G memory to process a single video at a time of resolution 856x480.
Software Dependencies No The paper states: 'We utilize Pytorch for our implementation.' However, it does not provide a specific version number for Pytorch or any other software dependencies.
Experiment Setup Yes We optimize the module (f RGB( ) and fα( )) weights using the entire test sequence with the Adam optimizer at a learning rate in the range of [0.00002, 0.002]... We use 100 epochs for a 60frame sequence in VOS and 60 epochs for dehazing and relighting... For getting a good performance on the UVOS task we utilize the following weights for the different losses; λrec = 1, λFsim = 0.001, λlayer = 1, λwarp = 0.01 and λMask = 0.01.