Retinomorphic Object Detection in Asynchronous Visual Streams
Authors: Jianing Li, Xiao Wang, Lin Zhu, Jia Li, Tiejun Huang, Yonghong Tian1332-1340
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experimental evaluation shows that our approach has significant improvements over the stateof-the-art methods with the single-modality, especially in high-speed motion and low-light scenarios. ... This section will first describe the experimental settings. Then, we conduct the effective test and ablation test to verify our approach. Finally, the scalability test provides quantitative results in various motion speeds and light conditions. |
| Researcher Affiliation | Academia | 1National Engineering Laboratory for Video Technology, School of Computer Science, Peking University, Beijing, China 2State Key Laboratory of Virtual Reality Technology and Systems, SCSE, Beihang University, Beijing, China 3Peng Cheng Laboratory, Shenzhen, China {lijianing, linzhu, tjhuang, yhtian}@pku.edu.cn, wangx03@pcl.ac.cn, jiali@buaa.edu.cn |
| Pseudocode | No | No explicit pseudocode or algorithm blocks are provided. |
| Open Source Code | No | The paper mentions an open-source simulator they used ("our open-source Vidar simulator (Kang et al. 2021)") and provides a link to their dataset ("Our dataset can be available at https://www.pkuml.org/resources/pku-vidar-dvs.html."), but it does not state that the source code for their proposed retinomorphic object detection methodology is released or available. |
| Open Datasets | Yes | Our dataset can be available at https://www.pkuml.org/resources/pku-vidar-dvs.html. ... To verify the effectiveness of our retinomorphic object detector, we conduct experiments on our newly built PKU-Vidar-DVS dataset and KITTI simulated dataset (Geiger, Lenz, and Urtasun 2012). |
| Dataset Splits | Yes | Afterward, we split them into three subsets for training, validation, and testing. ... This simulated dataset consists of 14 asynchronous visual streams for training, 3 hybrid streams (i.e., 0007, 0017, and 0018) for validating, and the remaining 3 hybrid streams (i.e., 0000, 0003, and 0006) for testing. |
| Hardware Specification | Yes | All networks are trained for 60 epochs with the Adam optimizer on an NVIDIA Tesla V100-PCLE GPU with the learning rate of 10-4. |
| Software Dependencies | No | The paper mentions software components like 'YOLOv3' and 'Adam optimizer' but does not specify their version numbers or the versions of other ancillary software such as programming languages (e.g., Python), deep learning frameworks (e.g., PyTorch, TensorFlow), or relevant libraries. |
| Experiment Setup | Yes | We set the overlap threshold to 0.5, the predicting score to 0.5 for Vidar, and 0.3 for DVS. ... All networks are trained for 60 epochs with the Adam optimizer on an NVIDIA Tesla V100-PCLE GPU with the learning rate of 10-4. We set λ to 10-3 for sparsity constrains in Equation (10) and the threshold θγ to 10-2 for dynamic interaction fusion in Equation (11). We set the temporal aggregation size T as 3 to make an accuracy-speed trade-off. |