Research map / Point clouds / Segmentation and detection
HEDNet: A Hierarchical Encoder-Decoder Network for 3D Object Detection in Point Clouds
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace SSR blocks with SED blocks (ablation) | HEDNet sparse backbone (no DED) · Waymo val 20% subset | L2 mAPH: 67.93 -> 69.89 | Helped |
| replace SSR with regular sparse residual blocks (ablation) | HEDNet sparse backbone · Waymo val 20% subset | L2 mAPH: 67.93 -> 68.30; latency 43 ms -> 176 ms | Mixed |
| add DED blocks (0 to 4) (ablation) | HEDNet · Waymo val 20% subset | L2 mAPH: 69.89 -> 71.92 | Helped |
| increase DED blocks from 4 to 5 (ablation) | HEDNet · Waymo val 20% subset | L2 mAPH: 71.92 -> 71.77 | Hurt |
| increase scales from 1 to 3 (encoder-decoder) (ablation) | HEDNet · Waymo val 20% subset | L2 mAPH: 69.88 -> 71.92 | Helped |
| increase scales from 3 to 4 (ablation) | HEDNet · Waymo val 20% subset | L2 mAPH: 71.92 -> 72.02; latency 67 ms -> 78 ms | No clear effect |
| stack 16 instead of 8 SSR blocks (ablation) | HEDNet-2D · Waymo val 20% subset | L2 mAPH: 67.80 -> 67.82 | No clear effect |
| replace SSR blocks with SED blocks (ablation) | HEDNet-2D · Waymo val 20% subset | L2 mAPH: 67.82 -> 69.42 | Helped |
| add second DED block in 2D variant (ablation) | HEDNet-2D · Waymo val 20% subset | L2 mAPH: 69.42 -> 69.67 | Helped |
| add DSVT training tricks (ablation) | VoxelNet · Waymo val 20% subset | L2 mAPH: 64.0 -> 69.1 | Helped |
| use smaller voxel size (ablation) | VoxelNet with tricks · Waymo val 20% subset | L2 mAPH: 69.1 -> 70.2 | Helped |
| replace backbone with HEDNet vs MinkUNet34-style | SemanticKITTI · SemanticKITTI val | mIoU: 68.3 -> 70.3 | Helped |
Closest papers by meaning
- SAFDNet: A Simple and Effective Network for Fully Sparse 3D Object Detection (CVPR 2024 · similarity 0.89)
- HVPR: Hybrid Voxel-Point Representation for Single-Stage 3D Object Detection (CVPR 2021 · similarity 0.89)
- VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection (CVPR 2018 · similarity 0.88)
- Point2Seq: Detecting 3D Objects As Sequences (CVPR 2022 · similarity 0.88)
- Uni3D: A Unified Baseline for Multi-Dataset 3D Object Detection (CVPR 2023 · similarity 0.88)
- Structure Aware Single-Stage 3D Object Detection From Point Cloud (CVPR 2020 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.