Spatiotemporal Deformation Perception for Fisheye Video Rectification
Authors: Shangrong Yang, Chunyu Lin, Kang Liao, Yao Zhao
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiment, Qualitative and Quantitative Comparison, Stability Analysis, Testing on Downstream Tasks, Ablation Study |
| Researcher Affiliation | Academia | Shangrong Yang, Chunyu Lin , Kang Liao, Yao Zhao Institute of Information Science, Beijing Jiaotong University Beijing Key Laboratory of Advanced Information Science and Network, Beijing, 100044, China {sr yang, cylin, kang liao, yzhao}@bjtu.edu.cn |
| Pseudocode | No | No explicit pseudocode or algorithm blocks were found. |
| Open Source Code | Yes | 1Available at https://github.com/uof1745-cmd/SDP |
| Open Datasets | Yes | We first use DAVIS (Perazzi et al. 2016) to synthesize our fisheye videos. DAVIS includes 90 videos for training and 30 for testing. Additionally, we apply Youtube-VOS (Xu et al. 2018) which contains 3400 training videos and 500 testing videos. |
| Dataset Splits | No | The paper specifies training and testing splits for DAVIS and Youtube-VOS datasets (e.g., 'DAVIS includes 90 videos for training and 30 for testing'; 'Youtube-VOS which contains 3400 training videos and 500 testing videos') but does not explicitly mention a 'validation' split. |
| Hardware Specification | Yes | We leverage the Adam optimizer with a learning rate of 1 10 4 to train the network on an NVIDIA Ge Force RTX 3090 for 30 epochs. |
| Software Dependencies | No | The paper mentions software components like 'Adam optimizer' and uses 'Raft (Teed and Deng 2020)' and 'VGG16 network', but does not provide specific version numbers for these or other relevant software dependencies. |
| Experiment Setup | Yes | In the experiment, we resize the input timestamp to 256 256 and set the batch size to 4. We leverage the Adam optimizer with a learning rate of 1 10 4 to train the network on an NVIDIA Ge Force RTX 3090 for 30 epochs. The overall loss is: L = λr LR + λa LA + λs LS + λf LF where {λr, λa, λs, λf} are the hyperparameter. We empirically set {20, 0.05, 250, 0.5}. |