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; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
UniDet3D: Multi-dataset Indoor 3D Object Detection
Authors: Maksim Kolodiazhnyi, Anna Vorontsova, Matvey Skripkin, Danila Rukhovich, Anton Konushin
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that Uni Det3D obtains significant gains over existing 3D object detection methods in 6 indoor benchmarks: Scan Net (+1.1 m AP50), S3DIS (+9.1 m AP50), ARKit Scenes (+19.4 m AP25), Multi Scan (+9.3 m AP50), 3RScan (+3.2 m AP50), and Scan Net++ (+2.7 m AP50). |
| Researcher Affiliation | Collaboration | 1Artificial Intelligence Research Institute, Moscow, Russia 2 NEURA Robotics Gmb H, Metzingen, Germany 3 University of Luxembourg, Luxembourg |
| Pseudocode | No | The paper describes the methodology in prose and mathematical formulas, but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states, "We implement Uni Det3D in the mmdetection3d (Contributors 2020) framework." It mentions using an existing framework and refers to baselines with publicly available code, but does not explicitly state that the source code for the Uni Det3D method itself is provided or made publicly available. |
| Open Datasets | Yes | We evaluate our method on six real-world indoor benchmarks: Scan Net (Dai et al. 2017), ARKit Scenes (Baruch et al. 2021), S3DIS (Armeni et al. 2016), Multi Scan (Mao et al. 2022), 3RScan (Wald et al. 2019), Scan Net++ (Yeshwanth et al. 2023). |
| Dataset Splits | Yes | Scan Net ... The training subset consists of 1201 scans, while 312 scans are used for validation. ARKit Scenes ... We use the official training and validation splits of 4493 and 549 scans, respectively. S3DIS ... We use the official Area 5 split, where 68 rooms serve for validation, and 204 rooms comprise the training subset. |
| Hardware Specification | Yes | All experiments are conducted using a single NVidia V100. |
| Software Dependencies | No | The paper mentions implementing Uni Det3D in the mmdetection3d (Contributors 2020) framework. While it refers to a specific framework with a publication year, it does not provide explicit version numbers for mmdetection3d or any other software libraries or dependencies used, which is required for a reproducible description. |
| Experiment Setup | Yes | All training details are the same as in One Former3D (Kolodiazhnyi et al. 2024), particularly, we use Adam W optimizer with an initial learning rate of 0.0001, weight decay of 0.05, batch size of 8, and polynomial scheduler with a base of 0.9 for 1024 epochs. We apply the standard augmentations: horizontal flipping, random rotations around the z-axis, and random scaling. During the training, we assign a ground truth object to the three nearest superpoints. Since during the inference we seek for oneto-one matching, we suppress redundant superpoints using NMS. No test-time augmentations are applied during the inference time. |