Research map / Label-efficient and robust learning / Recognition and long-tail
On the Provable Importance of Gradients for Autonomous Language-Assisted Image Clustering
Bo Peng, Jie Lu, Guangquan Zhang, Zhen Fang
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 |
|---|---|---|---|
| replace TAC filtering with GradNorm | CLIP ViT-B/32 clustering · ImageNet-Dogs | ACC: 75.1 -> 81.2 | Improved |
| replace TAC filtering with GradNorm | CLIP ViT-B/32 clustering · DTD | ACC: 45.9 -> 50.9 | Improved |
| replace TAC filtering with GradNorm | CLIP ViT-B/32 clustering · CIFAR-10 | ACC: 92.6 -> 91.1 | Hurt |
| replace TAC filtering with GradNorm | CLIP ViT-B/32 clustering · ImageNet-1K | ACC: 48.9 -> 52.6 | Improved |
| use stronger visual encoder ViT-L/14 instead of ViT-B/16 (ablation) | GradNorm · CIFAR-20 | ACC: 61.2 -> 61.7 | Improved |
| use GradNorm instead of TAC with ResNet-50 encoder (ablation) | CLIP ResNet-50 · CIFAR-10 | ACC: 69.5 -> 72.2 | Improved |
| use GradNorm instead of TAC with ViT-L/14 encoder (ablation) | CLIP ViT-L/14 · CIFAR-10 | ACC: 93.9 -> 95.3 | Improved |
| replace TAC with GradNorm on fine-grained data | CLIP ViT-B/32 · Food | ACC: 59.2 -> 67.8 | Improved |
| replace TAC with GradNorm under domain shift | CLIP ViT-B/32 · ImageNet-V2 | ACC: 38.0 -> 42.3 | Improved |
| increase number of selected positive nouns beta (ablation) | GradNorm · DTD and CIFAR-20 | clustering performance | Mixed |
| vary temperature kappa from 0.002 to 0.02 (ablation) | GradNorm · DTD and CIFAR-20 | clustering performance | No clear effect |
Closest papers by meaning
- Language-Assisted Image Clustering Guided by Discriminative Relational Signals and Adaptive Semantic Centers (NeurIPS 2026 · similarity 0.91)
- Image Clustering Conditioned on Text Criteria (ICLR 2024 · similarity 0.89)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.87)
- Target Semantics Clustering via Text Representations for Robust Universal Domain Adaptation (AAAI 2025 · similarity 0.87)
- Image Clustering with External Guidance (ICML 2024 · similarity 0.86)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (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.
