Research map / 3D vision / Depth and stereo
Implicit Event-RGBD Neural SLAM
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes EN-SLAM, an event-RGBD implicit neural SLAM that uses differentiable CRF rendering on a shared radiance field and event temporal aggregating optimization, plus two new event-RGBD datasets (DEV-Indoors, DEV-Reals).
Problem
Implicit neural SLAM fails under motion blur and lighting variation (convergence failures, tracking drift, distorted mapping) because RGB cameras have low dynamic range and low temporal resolution.
Approach
Model RGB and event cameras as two differentiable tone-mapping (CRF) decoders over one shared radiance/geometry field, supervise with event brightness differences rendered from camera poses, and use temporal event constraints with adaptive forward-query and probability-weighted ray sampling in tracking and global BA.
Key results
- On DEV-Indoors, EN-SLAM achieves average tracking ATE RMSE 8.56 cm vs 14.66 for ESLAM and 15.80 for CoSLAM, best in all 9 scenes.
- On DEV-Reals, average tracking ATE median 18.97 cm vs 30.95 for CoSLAM (second best).
- Runs at 17.4 FPS with 1.95M parameters, comparable to ESLAM's speed while being lighter.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use EN-SLAM instead of ESLAM | Neural SLAM · DEV-Indoors all avg | ATE RMSE (cm): 14.66 -> 8.56 | Helped |
| Use EN-SLAM instead of ESLAM | Neural SLAM · DEV-Indoors all avg | 3D accuracy (cm): 11.43 -> 9.06 | Helped |
| Use EN-SLAM instead of ESLAM | Neural SLAM · DEV-Indoors all avg | completion (cm): 9.35 -> 9.61 | Hurt |
| Use EN-SLAM instead of CoSLAM | Neural SLAM · DEV-Reals avg | ATE median (cm): 30.95 -> 18.97 | Helped |
| Add event tracking and event bundle adjustment (ablation) | EN-SLAM · #Rm blur, DEV-Indoors | ATE RMSE (cm): 18.75 -> 15.47 | Helped |
| Add event bundle adjustment (ablation) | EN-SLAM · #Rm blur, DEV-Indoors | ATE RMSE (cm): 17.07 -> 15.47 | Helped |
| Add event tracking (ablation) | EN-SLAM · #Rm blur, DEV-Indoors | ATE RMSE (cm): 19.72 -> 15.47 | Helped |
| Add differentiable CRF rendering (ablation) | EN-SLAM · #Dorm2, DEV-Reals | ATE RMSE (cm): 30.18 -> 15.47 | Helped |
| Add probability-weighted sampling (ablation) | EN-SLAM · #Rm blur, DEV-Indoors | ATE (cm): 9.86 -> 9.61 | Helped |
| Add probability-weighted sampling (ablation) | EN-SLAM · #Dorm2, DEV-Reals | ATE RMSE (cm): 19.78 -> 15.47 | Helped |
Surprising findings
- Removing the CRF module raised tracking ATE RMSE from 15.47 to 30.18 on DEV-Reals #Dorm2, indicating the imaging-model mismatch strongly hurts joint optimization.
Limitations
- Requires depth input, which may be unavailable in some scenarios.
- Focuses on indoor scenes and may struggle with boundless long trajectories.
Experimental setup
Datasets: DEV-Indoors DEV-Reals Vector
Benchmarks: DEV-Indoors DEV-Reals Vector
Metrics: ATE RMSE ATE median ATE mean Acc Comp Comp ratio Depth L1 PSNR SSIM LPIPS
Baselines: iMAP NICE-SLAM CoSLAM ESLAM EVO ESVO USLAM ORB-SLAM
Builds on: NICE-SLAM CoSLAM ESLAM Event generation model Camera response function
Closest papers by meaning
- Benchmarking Implicit Neural Representation and Geometric Rendering in Real-Time RGB-D SLAM (CVPR 2024 · similarity 0.86)
- Loopy-SLAM: Dense Neural SLAM with Loop Closures (CVPR 2024 · similarity 0.86)
- IBD-SLAM: Learning Image-Based Depth Fusion for Generalizable SLAM (CVPR 2024 · similarity 0.85)
- Photo-SLAM: Real-time Simultaneous Localization and Photorealistic Mapping for Monocular Stereo and RGB-D Cameras (CVPR 2024 · similarity 0.84)
- VTGaussian-SLAM: RGBD SLAM for Large Scale Scenes with Splatting View-Tied 3D Gaussians (ICML 2025 · similarity 0.84)
- SNI-SLAM: Semantic Neural Implicit SLAM (CVPR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.