Research map / Detection and segmentation / CNN features
FLSL: Feature-level Self-supervised Learning
Qing Su, Anton Netchaev, Hai Li, Shihao Ji
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 |
|---|---|---|---|
| increase gamma from 1 to 5 (ablation) | FLSL · bbox-aligned k-NN (ViT-S) | k-NN top-1: 66.6 -> 72.4 | Improved |
| remove Sinkhorn normalization (gamma=1) (ablation) | FLSL · bbox-aligned k-NN (ViT-S) | k-NN top-1: 71.2 -> 66.6 | Hurt |
| vary upsilon from 0.01 to 0.1 (ablation) | FLSL · bbox-aligned k-NN (ViT-S) | k-NN top-1: 68.7 -> 71.2 -> 65.1 | Mixed |
| increase number of centroids K from 1024 to 4096 (ablation) | FLSL · bbox-aligned k-NN | k-NN top-1: 68.1 -> 72.4 | Improved |
| increase K from 4096 to 16384 (ablation) | FLSL · bbox-aligned k-NN | k-NN top-1: 72.4 -> 72.1 | No clear effect |
| use FLSL instead of DINO pretraining | ViTDet-B/16 · COCO Mask R-CNN | AP bbox: 48.9 -> 53.1 | Improved |
| use FLSL instead of MAE pretraining | ViTDet-B/16 · COCO Mask R-CNN | AP mask: 45.5 -> 47.0 | Improved |
| use FLSL instead of ADCLR | ViT-S/16 Mask R-CNN · COCO | AP bbox: 44.3 -> 44.9 | Improved |
| use FLSL instead of DINO+SelfPatch | ViT-S/16 Mask R-CNN · COCO | AP bbox: 42.1 -> 44.9 | Improved |
| reduce patch size from 16 to 8 for FLSL | ViT-S Mask R-CNN · COCO | AP bbox: 44.9 -> 46.5 | Improved |
| use FLSL instead of DINO pretraining | ViT-S/8 Faster R-CNN · UAVDT | AP: 51.1 -> 55.2 | Improved |
| use FLSL instead of DINO pretraining | ViT-B/16 Faster R-CNN · UAVDT | AP: 49.1 -> 53.5 | Improved |
Later papers that compared with this method
1 later accepted paper reports FLSL as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Self-Supervised Learning of Intertwined Content and Positional Features for Object Detection (ICML 2025 · better on 2 of 2 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
- Patch-Level Representation Learning for Self-Supervised Vision Transformers (CVPR 2022 · similarity 0.88)
- Self-Supervised Learning of Object Parts for Semantic Segmentation (CVPR 2022 · similarity 0.87)
- A Survey on Self-Supervised Learning: Algorithms, Applications, and Future Trends (TPAMI 2024 · similarity 0.86)
- Self-Supervised Learning of Intertwined Content and Positional Features for Object Detection (ICML 2025 · similarity 0.86)
- Clustering Properties of Self-Supervised Learning (ICML 2025 · similarity 0.85)
- Self-Organizing Visual Prototypes for Non-Parametric Representation Learning (ICML 2025 · 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.
