Normal-NeRF: Ambiguity-Robust Normal Estimation for Highly Reflective Scenes

Authors: Ji Shi, Xianghua Ying, Ruohao Guo, Bowei Xing, Wenzhen Yue

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate that our method outperforms existing state-of-the-art methods on various datasets. Experiments Datasets. To comprehensively validate the effectiveness and robustness of our proposed method, we conduct evaluation on several datasets, including widely-used Ne RF Synthetic dataset (Mildenhall et al. 2021), two reflective objects datasets: Shiny Blender (Verbin et al. 2022) and Glossy Synthetic (Liu et al. 2023), and one real captured dataset from Ref-Ne RF (Verbin et al. 2022). Baselines and Metrics. We compare our method against the following baselines: Zip-Ne RF (Barron et al. 2023)... We evaluate rendering quality using PSNR, SSIM and LPIPS (Zhang et al. 2018), and assess normal accuracy with mean angular error (MAE) (Verbin et al. 2022). Ablation Studies. We conduct a series of ablation studies to evaluate the effect of our key components.
Researcher Affiliation Academia National Key Laboratory of General Artificial Intelligence School of Intelligence Science and Technology, Peking University EMAIL, EMAIL
Pseudocode No The paper describes methods verbally and mathematically but does not present any explicit pseudocode or algorithm blocks.
Open Source Code Yes Code https://github.com/sjj118/Normal-Ne RF
Open Datasets Yes To comprehensively validate the effectiveness and robustness of our proposed method, we conduct evaluation on several datasets, including widely-used Ne RF Synthetic dataset (Mildenhall et al. 2021), two reflective objects datasets: Shiny Blender (Verbin et al. 2022) and Glossy Synthetic (Liu et al. 2023), and one real captured dataset from Ref-Ne RF (Verbin et al. 2022).
Dataset Splits No The paper mentions evaluating on 'test views' of synthetic and real scenes but does not provide specific percentages, sample counts, or explicit references to how the datasets were split for training, validation, and testing within this work. It cites the datasets but not the specific split methodologies used.
Hardware Specification Yes All experiments are conducted on an NVIDIA RTX 4090 GPU.
Software Dependencies Yes We implement our model within Nerfstudio (Tancik et al. 2023), building upon the Instant-NGP (M uller et al. 2022) framework.
Experiment Setup Yes We train our model for 50k iterations with a batch size of 219 sample points. In all of our experiments, the parameter λn follows an exponential warmup, increasing from 0.01 to 1 over 20k iterations.