Research map / Graph learning / Knowledge and temporal graphs
LatentGNN: Learning Efficient Non-local Relations for Visual Recognition
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 |
|---|---|---|---|
| Add LatentGNN at Stages 3, 4, and 5 | ResNet-50(1x) Mask R-CNN · MSCOCO 2017 val | AP_box: 37.8 -> 39.5 | Helped |
| Add LatentGNN at Stages 3, 4, and 5 | ResNet-50(1x) Mask R-CNN · MSCOCO 2017 val | AP_sem: 34.2 -> 35.6 | Helped |
| Add one LatentGNN at Stage4 (ablation) | ResNet-50(1x) Mask R-CNN · MSCOCO 2017 val | AP_box: 37.8 -> 39.0 | Helped |
| Add one LatentGNN at Stage3 (ablation) | ResNet-50(1x) Mask R-CNN · MSCOCO 2017 val | AP_box: 37.8 -> 38.2 | Helped |
| Add one LatentGNN at Stage5 (ablation) | ResNet-50(1x) Mask R-CNN · MSCOCO 2017 val | AP_box: 37.8 -> 38.8 | Helped |
| Add three low-rank adjacency kernels instead of one (ablation) | Stage4 LatentGNN · MSCOCO 2017 val, ResNet-50(1x) | AP_box: 39.0 -> 39.2 | Helped |
| Add three low-rank adjacency kernels instead of one (ablation) | Stage4 LatentGNN · MSCOCO 2017 val, ResNet-50(1x) | AP_sem: 35.2 -> 35.4 | Helped |
| Add three low-rank adjacency kernels instead of one (ablation) | Stage345 LatentGNN · MSCOCO 2017 val, ResNet-50(1x) | AP_box: 39.5 -> 39.5 | No clear effect |
| Add three low-rank adjacency kernels instead of one (ablation) | LatentGNN · ScanNet test, PointNet++ Single Scale | Voxel Accuracy: 84.2 -> 85.2 | Helped |
| Use LatentGNN instead of a non-local block | ResNet-50(1x) Mask R-CNN, Stage4 · MSCOCO 2017 val | FLOPS overhead: +10.67G -> +1.11G | Helped |
| Use LatentGNN instead of a non-local block | PointNet++ · ScanNet | FLOPS overhead: +31M -> +15M | Helped |
Closest papers by meaning
- Representing Long-Range Context for Graph Neural Networks with Global Attention (NeurIPS 2021 · similarity 0.88)
- ClusterGNN: Cluster-Based Coarse-To-Fine Graph Neural Network for Efficient Feature Matching (CVPR 2022 · similarity 0.88)
- Beyond Grids: Learning Graph Representations for Visual Recognition (NeurIPS 2018 · similarity 0.88)
- Affinity Graph Supervision for Visual Recognition (CVPR 2020 · similarity 0.88)
- Spatial-Aware Graph Relation Network for Large-Scale Object Detection (CVPR 2019 · similarity 0.87)
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.87)
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.