Research map / Point clouds / Segmentation and detection
Learning Propagation for Arbitrarily-Structured Data
Sifei Liu, Xueting Li, Varun Jampani, Shalini De Mello, Jan Kautz
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add SGPN-prod propagation module | DRN-22-D · Cityscapes val, multi-scale | mIoU: 69.5 -> 75.0 | Improved |
| add SGPN-embed propagation module | Deeplab ResNet-38 · Cityscapes val, multi-scale | mIoU: 78.97 -> 80.42 | Improved |
| add SGPN-prod propagation module | Deeplab ResNet-38 · Cityscapes val, single-scale | mIoU: 78.20 -> 80.09 | Improved |
| propagate on superpixels with SGPN instead of pixels | DRN-22-D · Cityscapes val | mIoU: 69.5 -> 73.2 | Improved |
| replace SPN with SGPN (jointly learned edge representation) | DRN-22-D · Cityscapes val | mIoU: 71.1 -> 75.0 | Improved |
| replace non-local layer with SGPN propagation, same pairwise block | DRN-22-D · Cityscapes val | mIoU: 71.3 -> 75.0 | Improved |
| use SPN on a stronger Deeplab backbone | Deeplab ResNet-38 · Cityscapes val | mIoU | No clear effect |
| add joint 2D-3D training with 1D conv | DRN-22-D · RueMonge2014 point cloud | mIoU: 69.16 -> 70.35 | Improved |
| add SGPN propagation with image pairwise features and tangent-space DAG (ablation) | DRN-22-D · RueMonge2014 point cloud | mIoU: 70.35 -> 72.19 | Improved |
| add geometry-aware pairwise features (PG) to image features (PF) (ablation) | SGPN tangent-space DAG · RueMonge2014 point cloud | mIoU: 72.19 -> 73.66 | Improved |
| construct DAGs in tangent space instead of Euclidean space (ablation) | SGPN +PF+PG · RueMonge2014 point cloud | mIoU: 72.16 -> 73.66 | Improved |
| propagate globally over the whole point cloud instead of per image (ablation) | DRN image-to-points · RueMonge2014 point cloud | mIoU: 65.50 -> 69.16 | Improved |
Closest papers by meaning
- Parallel Sequence Modeling via Generalized Spatial Propagation Network (CVPR 2025 · similarity 0.87)
- Learning Affinity via Spatial Propagation Networks (NeurIPS 2017 · similarity 0.87)
- Boundary-Aware Feature Propagation for Scene Segmentation (ICCV 2019 · similarity 0.86)
- Universal Graph Convolutional Networks (NeurIPS 2021 · similarity 0.86)
- GSPN-2: Efficient Parallel Sequence Modeling (NeurIPS 2025 · similarity 0.85)
- LogoSP: Local-global Grouping of Superpoints for Unsupervised Semantic Segmentation of 3D Point Clouds (CVPR 2025 · similarity 0.84)
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.
