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..

DERD-Net: Learning Depth from Event-based Ray Densities

Authors: Diego de Oliveira Hitzges, Suman Ghosh, Guillermo Gallego

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on standard benchmarks (MVSEC and DSEC datasets) demonstrate unprecedented effectiveness: (i) using purely monocular data, our method achieves comparable results to existing stereo methods; (ii) when applied to stereo data, it strongly outperforms all state-of-the-art (SOTA) approaches, reducing the mean absolute error by at least 42%; (iii) our method also allows for increases in depth completeness by more than 3-fold while still yielding a reduction in median absolute error of at least 30%.
Researcher Affiliation Academia 1Technische Universität Berlin, Einstein Center Digital Future, Robotics Institute Germany 2Science of Intelligence Excellence Cluster, Germany
Pseudocode No The paper describes the methodology using textual descriptions and network architecture diagrams (Fig. 2), but it does not include a clearly labeled pseudocode or algorithm block.
Open Source Code Yes Project page: https://github.com/tub-rip/DERD-Net. We provide code, trained models and video results for clarity and reproducibility.
Open Datasets Yes Experiments on standard benchmarks (MVSEC and DSEC datasets) demonstrate unprecedented effectiveness: (i) using purely monocular data, our method achieves comparable results to existing stereo methods; (ii) when applied to stereo data, it strongly outperforms all state-of-the-art (SOTA) approaches, reducing the mean absolute error by at least 42%; (iii) our method also allows for increases in depth completeness by more than 3-fold while still yielding a reduction in median absolute error of at least 30%.
Dataset Splits Yes We employed three-fold cross-validation by utilizing two sequences for supervised training and reserving the remaining sequence for testing, repeating this process for all three possible combinations of sequences to ensure robustness in our evaluation. ... We select the commonly used Zurich_City_04_a sequence to provide a focused in-depth evaluation. We split the sequence into two halves for training and testing.
Hardware Specification Yes Our network achieved an average inference time of only 0.37 ms per Sub-DSI on an NVIDIA RTX A6000. ... DSI creation takes 45 ms, DSI fusion takes 26 ms, and pixel selection takes 0.2 ms on an 8-core computer with Intel Xeon(R) W-2225 CPU operating at 4.10 GHz.
Software Dependencies No The paper discusses the use of deep learning frameworks but does not specify particular software or library names with version numbers.
Experiment Setup Yes Table 2: Hyperparameters. Dataset Details: Batch 64, Optimizer AdamW, LR 10^-3, LF MAE, Epochs 3.