Research map / Label-efficient and robust learning / Self-supervised and contrastive
Franca: Nested Matryoshka Clustering for Scalable Visual Representation Learning
Shashanka Venkataramanan, Valentinos Pariza, Mohammadreza Salehi, Lukas Knobel, Elias Ramzi, Spyros Gidaris, Andrei Bursuc, Yuki M Asano
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 |
|---|---|---|---|
| Franca recipe (Matryoshka heads, CyclicMask, RASA) vs DINOv2 on same IN-21K data, ViT-L without HRFT | DINOv2 · ADE20K in-context | mIoU: 33.5 -> 34.5 | Improved |
| Franca vs DINOv2 on DAVIS VOS, ViT-L without HRFT/RASA | DINOv2 · DAVIS VOS | J&F: 64.2 -> 68.0 | Improved |
| Franca vs DINOv2 ViT-B kNN with HRFT+RASA | DINOv2 · IN-1K kNN | accuracy: 76.8 -> 79.6 | Improved |
| Franca (undistilled) vs official distilled DINOv2 ViT-B linear segmentation on VOC | DINOv2 (distilled) · VOC linear segmentation | mIoU: 90.2 -> 89.4 | Hurt |
| distill from ViT-G teacher (official DINOv2 vs IN-21K reproduction) ViT-B | DINOv2 · VOC linear segmentation | mIoU: 86.9 -> 90.2 | Improved |
| Franca ViT-G vs DINOv2-G in-context segmentation | DINOv2-G · Pascal VOC in-context | mIoU: 73.7 -> 76.7 | Improved |
| add RASA post-training to reduce positional bias (ablation) | Franca · patch cluster spatial entropy | mean spatial entropy | Improved |
| add Matryoshka representations, high-res finetune, RASA incrementally (Fig. 2) (ablation) | DINOv2-B IN-21K · IN-1K linear / VOC in-context | linear acc: 81.2 -> 82.6 | Improved |
| Franca vs DINOv2 on Feat2GS geometry probing | DINOv2 · Feat2GS Geometry | PSNR: 19.34 -> 19.58 | Improved |
| Franca vs SigLIP 2 on texture probing | SigLIP 2 · Feat2GS Texture | PSNR: 18.11 vs 17.97 | No clear effect |
Closest papers by meaning
- Hyperbolic Hierarchical Clustering for Visual Representation Learning (ECCV 2026 · similarity 0.87)
- ClusterFit: Improving Generalization of Visual Representations (CVPR 2020 · similarity 0.86)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.86)
- Neural Clustering based Visual Representation Learning (CVPR 2024 · similarity 0.85)
- Control-oriented Clustering of Visual Latent Representation (ICLR 2025 · similarity 0.85)
- Language-Assisted Image Clustering Guided by Discriminative Relational Signals and Adaptive Semantic Centers (NeurIPS 2026 · 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.
