Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Neural Atlas Graphs for Dynamic Scene Decomposition and Editing
Authors: Jan Philipp Schneider, Pratik S. Bisht, Ilya Chugunov, Andreas Kolb, Michael Moeller, Felix Heide
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Fit at test-time, NAGs achieve state-of-the-art quantitative results on the Waymo Open Dataset by 5 d B PSNR increase compared to existing methods and make environmental editing possible in high resolution and visual quality creating counterfactual driving scenarios with new backgrounds and edited vehicle appearance. We validate the proposed method on automotive scenes [36] and confirm that the method outperforms recent object-specific 3DGS baselines in visual quality by almost 5 d B PSNR on overall scene quality, and up to 11.2 d B PSNR for dynamic objects. |
| Researcher Affiliation | Collaboration | 1University of Siegen 2Princeton University 3Lamarr Institute 4Torc Robotics |
| Pseudocode | No | The paper describes methods using mathematical equations and descriptive text, but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Project Page: https://princeton-computational-imaging.github.io/nag/ and We release our code and all data on our code page. https://github.com/ jp-schneider/nag |
| Open Datasets | Yes | We rely on a subset of the Waymo [36] open dataset. We validate the generalization of our method on diverse outdoor scenes from the DAVIS dataset [30], which has been regularly used by matting methods [15, 24, 21] and provides high-resolution images (up to 1920 1080). |
| Dataset Splits | Yes | For evaluating the quality of our method in autonomous driving scenarios, we rely on a subset of the Waymo [36] open dataset. We select scenes with small ego motion but many objects, occlusions, and diverse and large object motions. In total, we evaluate on 7 scene segments including up to 199 images each, sampled at 10Hz using the forward camera feed. We divide each data segment into sequences of 21 to 89 images, leading to 25 subsequences to be reconstructed. |
| Hardware Specification | No | The main text of the paper does not specify the hardware used for experiments. The authors state in the NeurIPS checklist that this information is provided in the supplementary material. |
| Software Dependencies | No | The main text of the paper does not specify software dependencies with version numbers. The authors state in the NeurIPS checklist that code is released, implying dependencies would be part of that, but no versions are explicitly mentioned in the paper itself. |
| Experiment Setup | Yes | We jointly optimize the transformation parameters and network parameters of all nodes with the loss Latlas = ||ˆy y||1 + β ||ˆa m||1, which combines a photometric ℓ1 term with a mask loss that compares the predicted opacity ˆa to the input mask m M. The mask term encourages objects to remain opaque in masked regions while allowing flexibility in unmasked areas to represent shadows or object-induced effects. We empirically set β = 0.005 to balance these objectives. We weight the contributions of the networks using fixed scalars ηc = ηα = ηϕ = 0.1, and enforce valid ranges by clamping color values and applying a sigmoid to opacity, ensuring ci(x), αi(x) [0, 1]. |