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 [1].
Gaussian-Det: Learning Closed-Surface Gaussians for 3D Object Detection
Authors: Hongru Yan, Yu Zheng, Yueqi Duan
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on both synthetic and real-world datasets demonstrate that Gaussian-Det outperforms various existing approaches, in terms of both average precision and recall. To validate the effectiveness of the proposed Gaussian-Det, we have conducted experiments on both synthetic 3D-FRONT (Fu et al., 2021) and real-world Scan Net (Dai et al., 2017) datasets. The experimental results demonstrate that Gaussian-Det outperforms various existing approaches in terms of both average precision and recall. |
| Researcher Affiliation | Academia | Hongru Yan , Yu Zheng , Yueqi Duan Tsinghua University EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes methods and equations, but it does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block, nor does it present structured steps in a code-like format. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing source code, nor does it provide a link to a code repository. |
| Open Datasets | Yes | To validate the effectiveness of the proposed Gaussian-Det, we have conducted experiments on both synthetic 3D-FRONT (Fu et al., 2021) and real-world Scan Net (Dai et al., 2017) datasets. |
| Dataset Splits | No | The paper mentions using 3D-FRONT and Scan Net datasets and describes how some scenes for Scan Net were selected ("90 scenes that are randomly selected by (Hu et al., 2023)"), but it does not specify explicit training, validation, and test splits (e.g., percentages or exact counts) for their experiments. |
| Hardware Specification | Yes | We used a single NVIDIA RTX A6000 48GB GPU for both training and evaluation. |
| Software Dependencies | No | The paper mentions using Point Net++ as the backbone B and the official implementation of 3D-GS equipped with Su Ga R regularization. However, it does not provide specific version numbers for these or other software dependencies like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | The total training epochs of detection is 50 for both 3D-FRONT and Scan Net. For the input Gaussians G, we used the official implementation of 3D-GS (Kerbl et al., 2023) equipped with the Su Ga R regularization (Guédon and Lepetit, 2024), which takes 30,000 training iterations. Table 7: The hyper-parameter setup in Gaussian-Det. Name Description Value λres Weight of Lres 1 λpred Weight of Lpred 1 λrefine Weight of Lrefine 1 α Residue weight 0.1 δ Threshold of grouping the Gaussians 0.2 P The number of input Gaussians 40000 M The number of candidate Gaussians Gcand 1024 C Channel of candidate feature Fcand 256 |