Research map / Video / Video segmentation and tracking
GraphVid: It Only Takes a Few Nodes to Understand a Video
Eitan Kosman, Dotan Di Castro
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 |
|---|---|---|---|
| use relational edges (RGCN) instead of GCN (ablation) | GCN on video-graph · Kinetics-400 val (800 superpixels) | top-1 accuracy: 50.1 -> 66.2 | Improved |
| use relational edges (RGCN) instead of GAT (ablation) | GAT on video-graph · Kinetics-400 val (800 superpixels) | top-1 accuracy: 54.7 -> 66.2 | Improved |
| use RGCN instead of GAT for compute (ablation) | GAT on video-graph · Kinetics-400 val (800 superpixels) | FLOPs (x10^9): 56 -> 42 | Improved |
| add the full set of four augmentations (ablation) | RGCN-800 · Kinetics-400 val | top-1 accuracy: 66.2 -> 76.4 | Improved |
| add the full set of four augmentations (ablation) | RGCN-800 · Charades | mAP: 37.4 -> 43.1 | Improved |
| increase superpixels from 800 to 2000 | RGCN · Kinetics-400 val | top-1 accuracy: 76.4 -> 80.0 | Improved |
| pretrain on Kinetics-400 before Charades fine-tuning | RGCN-2000 · Charades | mAP: 45.3 -> 49.4 | Improved |
| add additive Gaussian edge noise (sigma 0.4) (ablation) | RGCN-800 · Kinetics-400 val | top-1 accuracy: 74.5 | Improved |
| add additive Gaussian node noise (sigma 0.2) (ablation) | RGCN-800 · Kinetics-400 val | top-1 accuracy: 73 | Improved |
| add random removal of spatial edges (ablation) | RGCN-800 · Kinetics-400 val | top-1 accuracy: 66 | Hurt |
| add random removal of superpixels (keep 0.8) (ablation) | RGCN-800 · Kinetics-400 val | top-1 accuracy: 70 | Improved |
| increase number of superpixels per frame (ablation) | GCN/GAT/RGCN · Kinetics-400 val | top-1 accuracy: saturates for S >= 650 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- AssembleNet-101 — AssembleNet: Searching for Multi-Stream Neural Connectivity in Video Architectures (ICLR 2020) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GCNext: Towards the Unity of Graph Convolutions for Human Motion Prediction (AAAI 2024 · similarity 0.87)
- Motion Graph Unleashed: A Novel Approach to Video Prediction (NeurIPS 2024 · similarity 0.87)
- LASER: A Neuro-Symbolic Framework for Learning Spatio-Temporal Scene Graphs with Weak Supervision (ICLR 2025 · similarity 0.86)
- DiffVsgg: Diffusion-Driven Online Video Scene Graph Generation (CVPR 2025 · similarity 0.86)
- Asynchronous Collaborative Graph Representation for Frames and Events (CVPR 2025 · similarity 0.85)
- HIG: Hierarchical Interlacement Graph Approach to Scene Graph Generation in Video Understanding (CVPR 2024 · 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.
