DPGS-SLAM: Diffusion-Prior and Provenance-Aware Monocular Gaussian Splatting SLAM for Dynamic Visual Sensing
Organizations: Graduate School of Information, Production and Systems, Waseda University, Kitakyushu, Fukuoka 808-0135, Japan · Faculty of Mechatronics, Warsaw University of Technology, 00-661 Warsaw, Poland · School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
Abstract
This paper presents DPGS-SLAM (Diffusion-Prior and Provenance-Aware Monocular Gaussian Splatting SLAM) for dense mapping in environments containing dynamic objects. Objects in dynamic motion remain a significant challenge for dense visual SLAM since dynamic observations can contaminate camera tracking and scene reconstruction. Although 3DGS (3D Gaussian splatting) enables efficient dense mapping, existing 3DGS-SLAM systems may exhibit instability when dynamic objects violate their underlying static-scene assumption. Excluding dynamic observations can reduce their influence on tracking and mapping, but may leave occluded static backgrounds missing or weakly constrained. Our proposed DPGS-SLAM is a monocular RGB SLAM framework that supports provisional representations of insufficiently observed static backgrounds through asynchronous diffusion-based background completion and provenance-aware Gaussian mapping. Dynamic regions are identified through optical-flow-based motion discovery and temporally consistent mask propagation by SAM2. For selected keyframes, a fine-tuned diffusion inpainting model generates plausible background appearance in occluded regions, while metric depth predicted from completed images initializes provisional Gaussian primitives. Provenance labels distinguish generated primitives from observation-supported primitives, excluding generated content from camera-pose estimation and observed-map optimization. When previously occluded backgrounds become visible, eligible generated primitives are removed, and observation-supported primitives are independently initialized from real RGB observations and predicted depth. Experimental results demonstrated competitive camera tracking, with average ATE RMSE values of 1.2 cm on TUM RGB-D and 2.7 cm on BONN RGB-D datasets. On StaDy4D, DPGS-SLAM improved average PSNR by 4.60 dB over WildGS-SLAM using only monocular RGB input.
Figures & tables
| Metrics | Scene | MonoGS (2024) [ 10 ] | WildGS (2025) [ 17 ] | DROID-W (2026) [ 35 ] | DPGS-SLAM (Ours) |
| PSNR [dB] | T0722_cctv | 16.6 | 18.9 | 21.8 | 25.2 |
| T1024_car | 16.3 | 19.9 | 19.7 | 24.5 | |
| T0316_ped | 12.6 | 24.8 | 24.0 | 27.7 | |
| SSIM [%] | T0722_cctv | 68.4 | 75.5 | 76.7 | 83.7 |
| T1024_car | 60.2 | 69.3 | 70.0 | 80.1 | |
| T0316_ped | 56.1 | 84.0 | 83.2 | 88.9 |
| Methods | Tracking [ms/frame] | Mapping [ms/frame] |
| MonoGS (2024) [ 10 ] | 579 | 1134 |
| WildGS-SLAM (2025) [ 17 ] | 372 | 3587 |
| DROID-W (2026) [ 35 ] | 249 | 1661 |
| DPGS-SLAM (Ours) | 510 | 1205 |
| Configuration | wk.xyz | wk.hf | wk.st | st.hf | Avg. |
| W/o motion masks | 2.6 | 3.4 | 2.1 | 4.2 | 3.1 |
| OCLR only | 2.0 | 3.1 | 1.3 | 3.0 | 2.4 |
| W/o geometric loss | 2.3 | 3.1 | 1.1 | 2.9 | 2.3 |
| Full DPGS-SLAM | 1.1 | 1.5 | 0.4 | 1.8 | 1.2 |
| Configuration | Inside mask | Outside mask | ||||
| PSNR [dB] | SSIM [%] | LPIPS ( ) | PSNR [dB] | SSIM [%] | LPIPS ( ) | |
| Hard mask | 11.5 | 43.7 | 50.5 | 25.2 | 80.2 | 23.6 |
| Soft mask ( ) | 13.9 | 45.4 | 51.0 | 25.0 | 81.4 | 22.7 |
| Adaptive weighting | 12.2 | 41.5 | 51.9 | 25.3 | 80.1 | 22.6 |
| W/o replacement | 16.5 | 56.2 | 43.5 | 25.9 | 82.7 | 22.0 |
| Full DPGS-SLAM | 22.1 | 73.7 | 36.3 | 26.1 | 84.8 | 21.4 |
| Configuration | Inside mask | Outside mask |
| Depth prediction | ||
| Metric3D v2 (static GT RGB) | 0.049 | 0.053 |
| Metric3D v2 (diffusion-completed RGB) | 0.053 | 0.051 |
| Gaussian map rendering | ||
| Hard mask | 0.523 | 0.087 |
| W/o replacement | 0.078 | 0.086 |