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..
Non-Line-of-Sight 3D Reconstruction with Radar
Authors: Haowen Lai, Zitong Lan, Mingmin Zhao
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our evaluation results demonstrate accurate and robust reconstruction in both LOS and NLOS regions. Code, dataset and demo videos are available on the project website. ... To evaluate our approach, we build a mobile robot prototype and collect data at 32 corners across 5 buildings, with 28k radar measurements in total. We train our models on 24 corners except 8 for testing. For evaluation, we measure the Chamfer distance, Hausdorff distance, and F-score between our predicted scenes and the ground truth. |
| Researcher Affiliation | Academia | Haowen Lai University of Pennsylvania EMAIL Zitong Lan University of Pennsylvania EMAIL Mingmin Zhao University of Pennsylvania EMAIL |
| Pseudocode | No | The paper describes methods and processes like ray tracing equations (2) and (3) and surface normal calculations (4) and (5), but it does not present these or any other procedures in a clearly labeled 'Pseudocode' or 'Algorithm' block with structured code-like formatting. |
| Open Source Code | No | Code, dataset and demo videos are available on the project website. (Abstract) ... Our code and dataset will be open-sourced after the review process. (NeurIPS Paper Checklist - Open access to data and code) |
| Open Datasets | No | Code, dataset and demo videos are available on the project website. (Abstract) ... Our code and dataset will be open-sourced after the review process. (NeurIPS Paper Checklist - Open access to data and code) |
| Dataset Splits | Yes | We collect a dataset from 32 distinct corners across 5 buildings... We use 24 corners for model training and the remaining 8 for evaluation. |
| Hardware Specification | Yes | All models are trained on an NVIDIA L40 GPU. ... All inferences are performed on an NVIDIA RTX 4070 GPU |
| Software Dependencies | No | All experiments use the Adam W optimizer, incorporating a warm-up period of 1k steps and an initial learning rate of 10 4 following a cosine annealing schedule. ... we adopt sparse 3D convolutions from Minkowski Engine [7]. |
| Experiment Setup | Yes | We trained this model using a batch size of 8 for 90k iterations. ... We train this model with a batch size of 4 for 60k iterations while freezing the first stage. All experiments use the Adam W optimizer, incorporating a warm-up period of 1k steps and an initial learning rate of 10 4 following a cosine annealing schedule. |