Accelerating the Training of Video Super-resolution Models
Authors: Lijian Lin, Xintao Wang, Zhongang Qi, Ying Shan
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that our method is capable of largely speeding up training (up to 6.2 speedup in wall-clock training time) without performance drop for various VSR models. |
| Researcher Affiliation | Collaboration | Lijian Lin, Xintao Wang*, Zhongang Qi, Ying Shan ARC Lab, Tencent PCG ljlin@stu.xmu.edu.cn, xintao.alpha@gmail.com, {zhongangqi, yingsshan}@tencent.com |
| Pseudocode | No | The paper describes methods in text and mathematical formulas but does not include any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code or links to a code repository. |
| Open Datasets | Yes | We conduct our experiments on the REDS (Nah et al. 2019) and Vimeo-90K (Xue et al. 2019b) datasets, which are widely-used and challenging datasets for VSR. |
| Dataset Splits | Yes | Following (Chan et al. 2021a; Wang et al. 2019), we adopt REDS4 as our test set, and use the left as training set. Vimeo-90K contains 64, 612 training, and 7, 824 testing 7-frame video sequences. |
| Hardware Specification | Yes | The training and analyses are performed with Py Torch on NVIDIA V100 GPUs in an internal cluster. |
| Software Dependencies | No | The paper mentions "Py Torch" as a framework but does not specify its version number or any other software dependencies with version numbers. |
| Experiment Setup | Yes | The spatial sizes in these two stages are set to max(32 32, H 2 ) and H W, respectively. The temporal sizes (i.e., number of consecutive frames fed into VSR models) in these three temporal stages are set in an increasing way: max(6, T 4 and T. ... The learning rate of training on 32 32 spatial size begins at 2e 4... we train these models with linear learning rate warmup for the first 5, 000 iterations. |