Research map / Detection and segmentation / Open-set and few-shot detection
TokenCut: Segmenting Objects in Images and Videos With Self-Supervised Transformer and Normalized Cut
Yangtao Wang, Xi Shen, Yuan Yuan, Yuming Du, Maomao Li, Shell Xu Hu, James L. Crowley, Dominique Vaufreydaz
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 CRF edge refinement (ablation) | TokenCut · ECSSD (saliency) | IoU (%): 71.2 -> 77.7 | Improved |
| add CRF edge refinement (ablation) | TokenCut (RAFT flow) · DAVIS (video) | Jaccard index: 64.3 -> 76.7 | Improved |
| use self-supervised DINO instead of supervised DeiT (ablation) | TokenCut · VOC07 (object discovery) | CorLoc: 2.39 -> 68.8 | Improved |
| use DINO instead of MoCoV3 (ablation) | TokenCut · VOC07 (object discovery) | CorLoc: 66.2 -> 68.8 | Improved |
| use DINO-B/16 instead of MAE-B/16 (ablation) | TokenCut · VOC07 (object discovery) | CorLoc: 61.5 -> 68.8 | Improved |
| use patch size 16 instead of 8 (ablation) | TokenCut · VOC07 (object discovery) | CorLoc: 67.3 -> 68.8 | No clear effect |
| increase tau from 0 to 0.3 (ablation) | TokenCut · DAVIS (video) | Jaccard index: 70.7 -> 76.7 | Improved |
| use mean split instead of EM (ablation) | TokenCut · VOC07 (object discovery) | CorLoc: 63.0 -> 68.8 | Improved |
| use mean split instead of K-means (ablation) | TokenCut · COCO20K | CorLoc: 58.8 -> 61.6 | Mixed |
| add optical flow to RGB in graph (ablation) | TokenCut · DAVIS (video, CRF) | Jaccard index: 62.2 -> 76.7 | Improved |
| add optical flow to RGB in graph (ablation) | TokenCut · FBMS (video, CRF) | Jaccard index: 67.5 -> 66.6 | Hurt |
| use transformer features of flow instead of mean flow (ablation) | TokenCut · DAVIS (video, CRF) | Jaccard index: 37.5 -> 76.7 | Improved |
Later papers that compared with this method
17 later accepted papers report TokenCut as a baseline in their results tables; 17 of them report a better number on most of the results they share with it.
- Learning Accurate Segmentation Purely from Self-Supervision (ECCV 2026 · better on 4 of 4 shared results)
- TriLite: Efficient Weakly Supervised Object Localization with Universal Visual Features and Tri-Region Disentanglement (CVPR 2026 · better on 2 of 2 shared results)
- Ensemble Foreground Management for Unsupervised Object Discovery (ICCV 2025 · better on 4 of 4 shared results)
- Beyond Single Images: Retrieval Self-Augmented Unsupervised Camouflaged Object Detection (ICCV 2025 · better on 1 of 1 shared result)
- Unsupervised Dense Prediction using Differentiable Normalized Cuts (ECCV 2024 · better on 4 of 4 shared results)
- CuVLER: Enhanced Unsupervised Object Discoveries through Exhaustive Self-Supervised Transformers (CVPR 2024 · better on 1 of 1 shared result)
- ProMerge: Prompt and Merge for Unsupervised Instance Segmentation (ECCV 2024 · better on 1 of 1 shared result)
- Pro2SAM: Mask Prompt to SAM with Grid Points for Weakly Supervised Object Localization (ECCV 2024 · better on 1 of 1 shared result)
9 more not listed.
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.
- DyStaB — DyStaB: Unsupervised Object Segmentation via Dynamic-Static Bootstrapping (CVPR 2021) · better on 0 of 3 shared results
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
- Self-Supervised Transformers for Unsupervised Object Discovery Using Normalized Cut (CVPR 2022 · similarity 0.92)
- Unsupervised Object Segmentation in Video by Efficient Selection of Highly Probable Positive Features (ICCV 2017 · similarity 0.88)
- CuVLER: Enhanced Unsupervised Object Discoveries through Exhaustive Self-Supervised Transformers (CVPR 2024 · similarity 0.88)
- Track and Segment: An Iterative Unsupervised Approach for Video Object Proposals (CVPR 2016 · similarity 0.87)
- Self-Supervised Human Detection and Segmentation via Background Inpainting (TPAMI 2021 · similarity 0.87)
- VideoCutLER: Surprisingly Simple Unsupervised Video Instance Segmentation (CVPR 2024 · similarity 0.86)
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.
