Research map / Detection and segmentation / CNN features
ClusterFomer: Clustering As A Universal Visual Learner
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 |
|---|---|---|---|
| Use ClusterFormer-Tiny instead of Swin-Tiny | Image classification · ImageNet-1K val | top-1 accuracy: 81.18% -> 81.31% | Helped |
| Use ClusterFormer-Small instead of Swin-Small | Image classification · ImageNet-1K val | top-1 accuracy: 83.02% -> 83.41% | Helped |
| Use ClusterFormer-Tiny instead of DINO with Swin-T | Object detection · COCO test-dev | mAP: 51.2 -> 52.0 | Helped |
| Use ClusterFormer-Small instead of DINO with Swin-S | Object detection · COCO test-dev | mAP: 53.3 -> 54.2 | Helped |
| Use ClusterFormer-Small instead of Mask2Former with Swin-S | Semantic segmentation · ADE20K val | mIoU: 51.1 -> 52.4 | Helped |
| Use ClusterFormer-Small instead of Mask-Dino with Swin-S | Instance segmentation · COCO test-dev | mAP: 46.5 -> 47.0 | Helped |
| Use ClusterFormer-Small instead of Mask-Dino with Swin-S | Panoptic segmentation · COCO Panoptic val | PQ: 54.9 -> 55.8 | Helped |
| Add recurrent cross-attention clustering to the baseline (ablation) | ClusterFormer-Tiny image classifier · ImageNet-1K validation | top-1 accuracy: 74.59% -> 80.57% | Helped |
| Add feature dispatching to the baseline (ablation) | ClusterFormer-Tiny image classifier · ImageNet-1K validation | top-1 accuracy: 74.59% -> 78.58% | Helped |
| Use recurrent cross-attention clustering instead of cosine similarity updating (ablation) | ClusterFormer-Tiny · ImageNet-1K validation | top-1 accuracy: 78.79 -> 81.31 | Helped |
| Use recurrent cross-attention clustering instead of K-Means (ablation) | ClusterFormer-Tiny · ImageNet-1K validation | top-1 accuracy: 80.96 -> 81.31 | Helped |
| Increase recursion iterations from 1 to 3 (ablation) | ClusterFormer-Tiny · ImageNet-1K validation | top-1 accuracy: 81.06% -> 81.31% | Helped |
Closest papers by meaning
- CLUSTSEG: Clustering for Universal Segmentation (ICML 2023 · similarity 0.89)
- Deep Ensemble Clustering for Visual Representation Learning (ICML 2026 · similarity 0.88)
- Unsupervised Universal Image Segmentation (CVPR 2024 · similarity 0.86)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.86)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.86)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 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.