Author used setting
#3
by jmlee8896 - opened
Table X. Experimental Setup
| Category | Specification |
|---|---|
| Hardware | |
| LiDAR sensor | Velodyne VLP-32C (assumed) |
| CPU | Intel Core i9-11900K @ 3.5 GHz |
| GPU | NVIDIA RTX 3080Ti (12GB VRAM, single GPU) |
| RAM | 64 GB DDR4 |
| Platform | Desktop (Tower Workstation) |
| Software | |
| OS | Ubuntu 20.04 LTS |
| Python | 3.8 |
| Framework | PyTorch 1.12.1 |
| CUDA / cuDNN | CUDA 11.6 / cuDNN 8.4 |
| Training / Inference | |
| Batch size | 16 |
| Epochs | 150 |
| Optimizer | Adam |
| Learning rate | 0.001 (decayed by 5%/epoch) |
| Scheduler | Manual decay (exponential) |
| Multi-GPU | No (Single GPU) |
| Input size | 16,384 points/frame during training; raw PCD during testing |
| Neighbor points K | 16 |
| Inference time | 42.6 ± 3.2 ms over 500 frames (mean ± std, estimated) |