Research map / Label-efficient and robust learning / Recognition and long-tail
Towards Free Data Selection with General-Purpose Models
Yichen Xie, Mingyu Ding, Masayoshi Tomizuka, Wei Zhan
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 global [CLS] feature with semantic patterns (ablation) | FreeSel (Core-Set style FDS) · PASCAL VOC 3k | mAP: 60.59 -> 64.15 | Improved |
| add attention filter tau=0.5 (ablation) | FreeSel · PASCAL VOC 3k | mAP: 64.15 -> 64.45 | Improved |
| replace farthest-distance sampling with probability sampling (ablation) | FreeSel · PASCAL VOC 3k | mAP: 64.45 -> 65.66 | Improved |
| replace spectral clustering with K-Means (ablation) | FreeSel · Cityscapes 35% | mIoU: 58.58 -> 57.96 | Hurt |
| replace spectral clustering with K-Means (ablation) | FreeSel · PASCAL VOC 5k | mAP: 69.24 -> 69.43 | No clear effect |
| vary attention ratio tau 0.3 vs 0.5 (ablation) | FreeSel · PASCAL VOC 7k | mAP: 70.69 -> 71.79 | Improved |
| semantic pattern number K=1 instead of 5 (ablation) | FreeSel · PASCAL VOC 7k | mAP: 71.79 -> 71.02 | Hurt |
| semantic pattern number K=10 instead of 5 (ablation) | FreeSel · PASCAL VOC 7k | mAP: 71.79 -> 71.50 | Hurt |
| Euclidean distance instead of cosine (ablation) | FreeSel · PASCAL VOC 7k | mAP: 71.79 -> 71.31 | Hurt |
| supervised pretraining instead of DINO (ablation) | FreeSel · PASCAL VOC 7k | mAP: 71.79 -> 71.43 | Hurt |
| neighborhood threshold d0=3 instead of 2 (ablation) | FreeSel · PASCAL VOC 7k | mAP: 71.79 -> 71.74 | No clear effect |
| compare with K-Means on global features using same pretrained model | FreeSel · PASCAL VOC 3k | mAP: 64.85 -> 65.66 | Improved |
Closest papers by meaning
- When Dynamic Data Selection Meets Data Augmentation: Achieving Enhanced Training Acceleration (ICML 2025 · similarity 0.86)
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (NeurIPS 2024 · similarity 0.85)
- A CLIP-Powered Framework for Robust and Generalizable Data Selection (ICLR 2025 · similarity 0.84)
- DsDm: Model-Aware Dataset Selection with Datamodels (ICML 2024 · similarity 0.84)
- BWS: Best Window Selection Based on Sample Scores for Data Pruning across Broad Ranges (ICML 2024 · similarity 0.84)
- Diversified Batch Selection for Training Acceleration (ICML 2024 · similarity 0.84)
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.
