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..
Dense Metric Depth Estimation via Event-based Differential Focus Volume Prompting
Authors: Boyu Li, Peiqi Duan, Zhaojun Huang, Xinyu Zhou, Yifei Xia, Boxin Shi
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Quantitative and qualitative results, including both in-domain and zero-shot experiments, demonstrate the superior performance of our method compared to existing approaches. |
| Researcher Affiliation | Academia | 1State Key Lab of Multimedia Info. Processing, School of Computer Science, Peking University 2National Eng. Research Center of Visual Tech., School of Computer Science, Peking University 3State Key Lab of General AI, School of Intelligence Science and Technology, Peking University EMAIL |
| Pseudocode | No | The paper describes the methodology in prose and uses diagrams (e.g., Figure 2 and 3) to illustrate the pipeline and module structures, but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | Code and data will be available at https://github.com/liboyu02/EDFV/. Both the dataset and the codes will be released upon acceptance. |
| Open Datasets | Yes | We further construct synthetic and real-captured datasets to facilitate the training and evaluation of both frame-based and event-based methods. ... These datasets will be made public to facilitate future research. ... We further capture a semi-real dataset using 4D Light Field Dataset [21] as the source. |
| Dataset Splits | Yes | We select 150 scenes as training data and 60 scenes as test data, containing AIF images, GT depth maps, events and focal stacks. ... Our training set contains 130 scenes and test set contains 30 scenes. ... For all the datasets, we select 5 frames with uniformly split focal depths as the input of frame-based DFF methods for training and evaluation... |
| Hardware Specification | Yes | We implement our method using the Pytorch framework and run on a single NVIDIA Ge Force RTX 4090 GPU. |
| Software Dependencies | No | We implement our method using the Pytorch framework and run on a single NVIDIA Ge Force RTX 4090 GPU. We use Adam W [36] optimizer in the training phrase. |
| Experiment Setup | Yes | For each dataset, we train for 700 epochs with initial learning rate 3e-4 and weight decay 1e-5. We randomly crop the input images into 320x640 resolution with random flipping afterwards. For loss functions, we set α = 0.1, β0 = 1, β1 = β2 = β3 = 0.1 in all our experiments, γ = 5 for Blender-Syn dataset and γ = 1 for Sintel-Dr. Bokeh. |