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..
Binary Radiance Fields
Authors: Seungjoo Shin, Jaesik Park
NeurIPS 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In our experiments, binary radiance field representation successfully outperforms the reconstruction performance of state-of-the-art (SOTA) storage-efficient radiance field models with lower storage allocation. |
| Researcher Affiliation | Academia | Seungjoo Shin GSAI, POSTECH EMAIL Jaesik Park CSE & IPAI, Seoul National University EMAIL |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper provides a project homepage URL (https://seungjooshin.github.io/Bi RF) but does not include an explicit statement about releasing source code or a direct link to a code repository. |
| Open Datasets | Yes | We use two synthetic datasets: the Synthetic-Ne RF dataset [1] and the Synthetic-NSVF dataset [13]... We also employ the Tanks and Temples dataset [41]... |
| Dataset Splits | No | The paper mentions '100 training views and 200 test views' for the Synthetic-Ne RF and Synthetic-NSVF datasets, and 'training and test views' for Tanks and Temples, but does not explicitly specify a validation split or its size. |
| Hardware Specification | Yes | We optimize all our models for 20K iterations on a single GPU (NVIDIA RTX A6000). |
| Software Dependencies | No | The paper mentions using 'Instant-NGP [20]', 'Nerf Acc [42]', and 'Adam [43] optimizer' but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | We use 16 levels of multi-resolution 3D feature grids with resolutions from 16 to 1024, while each grid includes up to T3D feature vectors. We also utilize four levels of multi-resolution 2D feature grids with resolutions from 64 to 512, while each grid includes up to T2D feature vectors... We use the Adam [43] optimizer with an initial learning rate of 0.01, which we decay at 15K and 18K iterations by a factor of 0.33. Furthermore, we adopt a warm-up stage during the first 1K iterations to ensure stable optimization. We set λsparsity = 2.0 10 5 in this work. |