Research map / Video / Video segmentation and tracking
Iterative Knowledge Exchange Between Deep Learning and Space-Time Spectral Clustering for Unsupervised Segmentation in Videos
Emanuela Haller, Adina Magda Florea, Marius Leordeanu
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 RAFT instead of FlowNet2.0 for graph motion and features (ablation) | Graph Module (unsupervised) · DAVIS | J Mean: 68.1 -> 70.7 | Improved |
| concatenate FlowNet2.0 features with RAFT features in the node features (ablation) | Graph Module (RAFT graph) · DAVIS | J Mean: 70.7 -> 70.8 | Improved |
| concatenate RAFT features onto FlowNet2.0 graph (ablation) | Graph Module (FlowNet2.0 graph) · DAVIS | J Mean: 68.1 -> 70.6 | Improved |
| add graph module space-time clustering on FCN features (RAFT) | FCN backbone · DAVIS | J Mean: 66.5 -> 78.2 | Improved |
| add graph module on DeepLabv3 features (RAFT) | DeepLabv3 backbone · DAVIS | J Mean: 67.0 -> 77.7 | Improved |
| add graph module on 3DC-Seg features (RAFT) | 3DC-Seg · DAVIS | J Mean: 84.3 -> 85.9 | Improved |
| add graph module on MATNet features (RAFT) | MATNet · DAVIS | J Mean: 82.4 -> 84.5 | Improved |
| add graph module on AnDiff features (RAFT) | AnDiff · DAVIS | J Mean: 81.7 -> 83.7 | Improved |
| add graph module on COSNet features (RAFT) | COSNet · DAVIS | J Mean: 80.5 -> 83.4 | Improved |
| add Network Module to unsupervised IKE (graph cycle 1 -> 2) (ablation) | IKE graph module · DAVSOD, DAVIS, YouTube-Objects, SegTrack | relative change: 1.3% -> 10.4% | Improved |
| add a third graph-network cycle (unsupervised, with network) (ablation) | IKE graph module · four datasets | relative change: 10.4% -> 1.7% | Improved |
| add Network Module when starting from supervised DeepLabv3 features (ablation) | IKE · four datasets | relative change: 2.2% -> 1.0% | Improved |
Closest papers by meaning
- Deep Spectral Methods: A Surprisingly Strong Baseline for Unsupervised Semantic Segmentation and Localization (CVPR 2022 · similarity 0.90)
- Temporally-Weighted Hierarchical Clustering for Unsupervised Action Segmentation (CVPR 2021 · similarity 0.89)
- Learning Neural Eigenfunctions for Unsupervised Semantic Segmentation (ICCV 2023 · similarity 0.89)
- UniViT: Unifying Image and Video Understanding in One Vision Encoder (NeurIPS 2025 · similarity 0.88)
- Self-supervised Object-Centric Learning for Videos (NeurIPS 2023 · similarity 0.88)
- Unsupervised Learning of Foreground Object Segmentation (IJCV 2019 · 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.
