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..
NFL-BA: Near-Field Light Bundle Adjustment for SLAM in Dynamic Lighting
Authors: Andrea Dunn Beltran, Daniel Rho, Marc Niethammer, Roni Sengupta
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our evaluations mainly focus on endoscopy procedure where SLAM can enable autonomous navigation, guidance to unsurveyed regions, blindspot detections, and 3D visualizations, which can significantly improve patient outcomes and endoscopy experience for both physicians and patients. Replacing Photometric Bundle Adjustment loss of SLAM systems with NFL-BA leads to significant improvement in camera tracking, 37% for Mono GS and 14% for Endo GS, and leads to state-of-the-art camera tracking and mapping performance on the C3VD colonoscopy dataset. Further evaluation on indoor scenes captured with phone camera with flashlight turned on, also demonstrate significant improvement in SLAM performance due to NFL-BA. |
| Researcher Affiliation | Academia | 1 University of North Carolina at Chapel Hill 2 University of California San Diego |
| Pseudocode | No | The paper describes methods and equations but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | We are willing to make captured data and code publicly available once accepted. |
| Open Datasets | Yes | C3VD. The C3VD dataset [4] (CC BY-NC-SA 4.0) was created using a phantom colon with synthetic materials to simulate realistic tissue geometry. Colon10K. To test generalization in realworld clinical endoscopy settings, we evaluate on Colon10K [24], a large-scale video dataset without depth or pose supervision. |
| Dataset Splits | No | We choose these 8 videos from the test split of PPSNet to avoid any bias when the SLAM is initialized with PPSNet predicted depth map. The names of the sequences are as follows: cecum_t1_a, cecum_t2_a, cecum_t3_a, sigmoid_t3_a, desc_t4_a_p2, trans_t2_a, trans_t3_a, and trans_t4_a. ... We also evaluate on a subset of 8 videos with at least one video from each section of the colon, with varying camera motion, and anomalies. |
| Hardware Specification | Yes | We trained all models on a single NVIDIA RTX A6000 GPU. |
| Software Dependencies | No | The paper does not explicitly state specific software dependencies with version numbers in the main text or appendices. |
| Experiment Setup | Yes | Because NFL-BA is designed as a drop-in replacement for photometric bundle adjustment, we only adjusted the two associated loss weights; all other hyperparameters remain identical between the Photo-BA and NFL-BA experiments. Please see supplemental for detailed hyperparameter settings. ... We also used the default loss weights of NICE-SLAM, setting λren to 0.5 during tracking and 0.2 during mapping, and λgeo to 1 in both phases. ... We used the default loss weights of Endo GSLAM, λren; λgeo, set to 0.5 and 1 during tracking and 1 and 1 during mapping, respectively. ... For all input depths, we set λren and λgeo to 0.8 and 0.5, respectively. |