DENSE RGB SLAM WITH NEURAL IMPLICIT MAPS
Authors: Heng Li, Xiaodong Gu, Weihao Yuan, luwei yang, Zilong Dong, Ping Tan
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our method on commonly used benchmarks and compare it with modern RGB and RGB-D SLAM systems. Our method achieves favorable results than previous methods and even surpasses some recent RGB-D SLAM methods. |
| Researcher Affiliation | Collaboration | 1Hong Kong University of Science and Technology, 2Alibaba Group, 3Simon Fraser University |
| Pseudocode | No | The paper does not contain any pseudocode or algorithm blocks. It describes the pipeline textually and with a diagram. |
| Open Source Code | Yes | The code is at poptree.github.io/DIM-SLAM/. |
| Open Datasets | Yes | We evaluate our method on three widely used datasets: TUM RGB-D (Sturm et al., 2012a), Eu Ro C (Burri et al., 2016), and Replica (Straub et al., 2019) dataset. |
| Dataset Splits | No | The paper mentions using datasets for evaluation and provides details on sampling points for rendering and optimization windowing, but it does not specify explicit training/validation/test dataset splits with percentages, sample counts, or predefined split references. |
| Hardware Specification | Yes | Our method is implemented with Py Torch (Paszke et al., 2017) and runs on a server with two NVIDIA 2080Ti GPUs. |
| Software Dependencies | No | The paper mentions that the method is implemented with PyTorch but does not provide specific version numbers for PyTorch or any other software dependencies required for reproducibility. |
| Experiment Setup | Yes | We fix the iteration number of initialization Ni to 1500, the number of sampling pixels |M| to 3000, the size of the window |W| to 21. The iteration number of window optimization for ours(one thread) Nw is 100. In our two-thread version, we change Nw to 20 for tracking. We use the Adam (Kingma & Ba, 2014) optimizer to optimize both the implicit map representation and the camera poses, with learning rates of 0.001 and 0.01, respectively. The αwarping, αrender, αsmooth is 0.5, 0.1, 0.01, respectively. |