Unblur-SLAM: Dense Neural SLAM for Blurry Inputs
About
We propose Unblur-SLAM, a novel RGB SLAM pipeline for sharp 3D reconstruction from blurred image inputs. In contrast to previous work, our approach is able to handle different types of blur and demonstrates state-of-the-art performance in the presence of both motion blur and defocus blur. Moreover, we adjust the computation effort with the amount of blur in the input image. As a first stage, our method uses a feed-forward image deblurring model for which we propose a suitable training scheme that can improve both tracking and mapping modules. Frames that are successfully deblurred by the feed-forward network obtain refined poses and depth through local-global multi-view optimization and loop closure. Frames that fail the first stage deblurring are directly modeled through the global 3DGS representation and an additional blur network to model multiple blurred sub-frames and simulate the blur formation process in 3D space, thereby learning sharp details and refined sub-frame poses. Experiments on several real-world datasets demonstrate consistent improvements in both pose estimation and sharp reconstruction results of geometry and texture.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| 3D Scene Deblurring | Deblur-NeRF motion blur and defocus blur | PSNR29.49 | 8 | |
| 3D Scene Deblurring | Deblur-NeRF defocus blur | PSNR27.45 | 6 | |
| Trajectory Estimation | TUM | ATE RMSE0.336 | 4 | |
| Trajectory Estimation | MCD | ATE RMSE0.128 | 4 | |
| SLAM | TUM RGBD (fr1_desk, fr2_xyz, fr3_office) | FPS0.85 | 4 | |
| SLAM | ArchViz-2 | ATE0.0027 | 2 | |
| SLAM | ArchViz 3 | ATE0.0067 | 2 | |
| SLAM | ArchViz | Average Translational Error0.0056 | 2 | |
| SLAM | ArchViz 1 | ATE0.0075 | 2 |