AS-Det: Active Sampling for Adaptive 3D Object Detection in Point Clouds
Authors: Ziheng Ding, Xiaze Zhang, Qi Jing, Ying Cheng, Rui Feng
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the superior performance and adaptability of AS-Det on both Li DAR and 4D Radar point clouds. Experiments conducted on various point clouds showcase that AS-Det outperforms other methods on both Li DAR and 4D Radar benchmarks, demonstrating the effectiveness and adaptability of our approach. The paper includes a dedicated section '4 Experimental Analysis' with subsections '4.1 Benchmarks', '4.2 Results on Li DAR Benchmarks', '4.3 Results on 4D Radar Benchmarks', and '4.4 Ablation Studies', featuring tables with performance metrics like m AP and NDS. |
| Researcher Affiliation | Academia | All authors are affiliated with 'Fudan University', as indicated by '1School of Computer Science, Shanghai Key Laboratory of Intelligent Information Processing, Fudan University' and email domains like 'm.fudan.edu.cn' and 'fudan.edu.cn'. This indicates an academic affiliation. |
| Pseudocode | No | The paper describes the 'Pipeline' of the Active Sampling strategy in Section 3.2 with numbered steps (1) Value Prediction, (2) Normalization, and (3) Sampling. However, these steps are described in a narrative and equation-based format, not as a formally structured pseudocode block or algorithm. |
| Open Source Code | Yes | Code https://github.com/eat-slim/AS-Det |
| Open Datasets | Yes | (a) KITTI (Geiger, Lenz, and Urtasun 2012) is a widely used Li DAR benchmark... (b) nu Scenes (Caesar et al. 2019) is a 360-degree FOV Li DAR benchmark... (c) Vo D (Palffy et al. 2022) is a pioneering benchmark of 4D Radar... |
| Dataset Splits | Yes | Comparison among point-based and single-stage voxel-based methods on the KITTI test split set with APR40. Comparison on the nu Scenes (Li DAR point cloud) val split set. Comparison on the Vo D (4D Radar point cloud) val split set with APR40. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used to conduct the experiments. It mentions Li DAR and 4D Radar as sensor types but not computational hardware. |
| Software Dependencies | No | The paper does not explicitly state specific version numbers for any software dependencies, libraries, or frameworks used in the implementation or experimentation (e.g., Python version, PyTorch version, CUDA version). |
| Experiment Setup | No | The paper describes the overall loss function (L = Lsample + Lcenter + Ldetection) and the types of losses used (smooth-l1, BCE), but it does not provide concrete hyperparameter values such as learning rate, batch size, number of epochs, or the specific optimizer configuration used for training the model. |