MetaNeRV: Meta Neural Representations for Videos with Spatial-Temporal Guidance

Authors: Jialong Guo, Ke Liu, Jiangchao Yao, Zhihua Wang, Jiajun Bu, Haishuai Wang

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments conducted on multiple datasets demonstrate the superiority of Meta Ne RV for video representations and video compression. Experimental results on multiple datasets demonstrate that Meta Ne RV outperforms other frame-wise methods in both video representations. We conduct quantitative and qualitative comparison experiments on 8 different video datasets to evaluate our Meta Ne RV against Ne RV-based methods for video representation tasks.
Researcher Affiliation Academia 1Zhejiang Key Laboratory of Accessible Perception and Intelligent Systems, College of Computer Science, Zhejiang University, China 2Cooperative Medianet Innovation Center, Shanghai Jiaotong University, China EMAIL, EMAIL
Pseudocode No The paper describes the methods in paragraph text and refers to 'Further details regarding the algorithm can be found in Appendix Section B.' but does not present any structured pseudocode or algorithm blocks within the provided text.
Open Source Code Yes code https://github.com/jialong2023/Meta Ne RV
Open Datasets Yes The datasets include multiple real-world datasets across various video types, such as UCF101(Soomro, Zamir, and Shah 2012), HMDB51(Kuehne et al. 2011), and MCL JCV (Wang et al. 2016), as well as ultrasound datasets like Echo CP (Wang et al. 2021), and Echo Net-LVH (Duffy et al. 2022). Furthermore, we conduct inference experiments on HOLLYWOOD2 (Marcin 2009), SVW (Safdarnejad et al. 2015), and OOPS (Epstein, Chen, and Vondrick 2020).
Dataset Splits Yes We selected 900 videos for each dataset, 800 for the training set, and 100 for the test set.
Hardware Specification Yes We conduct all experiments with RTXA6000 GPU, while the number of inner loop steps is 3.
Software Dependencies No The paper mentions 'Pytorch' and 'Adam optimizer' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes We set up-scale factors 5, 2, 2, 2, 2 for each block of our Meta Ne RV model to reconstruct a 320 240 image from the feature map of size 4 3. We train the model using Adam optimizer (Kingma and Ba 2014) with a learning rate 1e4 by Pytorch. We conduct all experiments with RTXA6000 GPU, while the number of inner loop steps is 3.