Better and Faster: Adaptive Event Conversion for Event-Based Object Detection
Authors: Yansong Peng, Yueyi Zhang, Peilin Xiao, Xiaoyan Sun, Feng Wu
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on three event-based detection datasets (1Mpx, Gen1, and MVSEC-NIGHTL21) demonstrate that our proposed approach outperforms other state-of-the-art methods by a large margin, while achieving a much faster running speed (< 14 ms and < 4 ms for 50 ms event data on the 1Mpx and Gen1 datasets). |
| Researcher Affiliation | Academia | University of Science and Technology of China, Hefei, China, 230026 {pengyansong, peilinxiao}@mail.ustc.edu.cn, {zhyuey, sunxiaoyan, fengwu}@ustc.edu.cn |
| Pseudocode | Yes | Algorithm 1: Hyper Histogram Generation in AEC Module |
| Open Source Code | No | The paper does not provide any specific link or statement about making the source code for its methodology publicly available. |
| Open Datasets | Yes | We evaluate our proposed methods on three representative datasets, which are the 1Mpx Detection dataset (Perot et al. 2020), the Gen1 Detection dataset (de Tournemire et al. 2020) and the Nighttime Driving Detection dataset MVSECNIGHTL21 (Hu, Liu, and Delbruck 2021). |
| Dataset Splits | Yes | The 1Mpx Detection dataset contains a total of 14.65 hours of events, of which 11.19 hours are for training, 2.21 hours are for validation, and 2.25 hours are for testing. The Gen1 Detection dataset... consists of 39 hours of events, of which 22.63 hours are for training, 6.59 hours for validation, and 10.10 hours for testing. |
| Hardware Specification | Yes | When training the models, we utilize 8 NVIDIA GTX 1080Ti GPUs for 100 epochs. When testing, only one NVIDIA GTX 1080Ti GPU is used. |
| Software Dependencies | No | The paper mentions using YOLOv5, Deformable-DETR, and Retina Net, but does not provide specific version numbers for these frameworks or any other software dependencies like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | For fair comparisons, we choose the most commonly used 50 ms as the time interval. The accumulated time T of the queue in AEC is also set to 50 ms. C and ts are set as 4 and 50 ms / 8... The values of the threshold L on three datasets are chosen as 100000, 10000, and 5750 respectively. |