Research map / Detection and segmentation / CNN features
Probabilistic Graph Attention Network with Conditional Kernels for Pixel-Wise Prediction
Dan Xu, Xavier Alameda-Pineda, Wanli Ouyang, Elisa Ricci, Xiaogang Wang, Nicu Sebe
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 attention-gated CRFs (FLAG) instead of removing AG-CRFs (ablation) | PGA-Net · NYUD-V2 RGB | ODS: .722 -> .744 | Improved |
| use attention-gated CRFs instead of plain CRFs (ablation) | PGA-Net · Pascal-Context | mIoU: 49.12 -> 51.01 | Improved |
| use fully learned FLAG instead of partially learned PLAG gates (ablation) | PGA-Net · NYUD-V2 RGB | AP: .746 -> .765 | Improved |
| add conditional kernels (CK) (ablation) | PGA-Net FLAG-CRFs · NYUD-V2 RGB | AP: .765 -> .778 | Improved |
| add conditional kernels (CK) (ablation) | PGA-Net FLAG-CRFs · Pascal-Context | mIoU: 51.01 -> 52.15 | Improved |
| add deep supervision (ablation) | PGA-Net · Pascal-Context | mIoU: 48.92 -> 51.01 | Improved |
| add first-level hierarchy to baseline (H2) (ablation) | PGA-Net · NYUD-V2 RGB | ODS: .711 -> .722 | Improved |
| use five scales instead of three (ablation) | PGA-Net w/ FLAG-CRFs + CK · Pascal-Context | mIoU: 52.15 -> 53.25 | Improved |
| use five scales instead of three (ablation) | PGA-Net w/ CK · BSDS500 | ODS: .799 -> .805 | Improved |
| use HRNetV2-W48 backbone instead of ResNet50 (4S) | PGA-Net · Pascal-Context | mIoU: 53.25 -> 55.18 | Improved |
| add FLAG-CRFs and CK to baseline (ablation) | PGA-Net · KITTI depth | rel: 0.152 -> 0.118 | Improved |
| add DORN to PGA-Net | PGA-Net w/ FLAG-CRFs + CK · KITTI depth | rmse: 4.449 -> 2.634 | Improved |
Closest papers by meaning
- Learning Deep Structured Multi-Scale Features using Attention-Gated CRFs for Contour Prediction (NeurIPS 2017 · similarity 0.89)
- Convolutional Kernel Networks for Graph-Structured Data (ICML 2020 · similarity 0.87)
- Structured Attention Guided Convolutional Neural Fields for Monocular Depth Estimation (CVPR 2018 · similarity 0.87)
- Attention to Scale: Scale-Aware Semantic Image Segmentation (CVPR 2016 · similarity 0.86)
- GraphFPN: Graph Feature Pyramid Network for Object Detection (ICCV 2021 · similarity 0.86)
- Graph Attention is Not Always Beneficial: A Theoretical Analysis of Graph Attention Mechanisms via Contextual Stochastic Block Models (ICML 2025 · similarity 0.85)
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.
