Learning Scale-Aware Spatio-temporal Implicit Representation for Event-based Motion Deblurring

Authors: Wei Yu, Jianing Li, Shengping Zhang, Xiangyang Ji

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments demonstrate that our SASNet outperforms stateof-the-art methods on both synthetic Go Pro and real H2D datasets, especially in high-speed motion scenarios.
Researcher Affiliation Academia 1School of Computer Science and Technology, Harbin Institute of Technology, Weihai, China 2School of Computer Science, Peking University, Beijing, China. 3Department of Automation, Tsinghua University, Beijing, China.
Pseudocode No The paper provides figures of network architectures (Figure 2, 3, 4) and describes the method in text, but no explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code and dataset are available at https://github.com/aipixel/SASNet.
Open Datasets Yes Code and dataset are available at https://github.com/aipixel/SASNet.
Dataset Splits Yes Go Pro Dataset. It consists of 3214 sharp images with resolutions of 1280 720, in which 2103 are used for training and 1111 for testing.
Hardware Specification Yes The proposed SASNet is implemented by Py Torch and trained on an NVIDIA Ge Force RTX 3090 for 100 epochs with 8 batch sizes.
Software Dependencies No The proposed SASNet is implemented by Py Torch... In Py Torch (Paszke et al., 2019)...
Experiment Setup Yes The training patch size is set to 256 256 and augmented by horizontal and vertical flipping to enhance its robustness. We use the Adam optimizer (Kingma & Ba, 2014) with an initial learning rate of 10 4 that linear decays by 0.5 for every 30 epoch and only employ L1 loss as the training loss.