ReImageNet Reannotates 50,000 ImageNet Val Images With 96,051 Boxes, New Class Definitions
ducha_aiki · x · 2026-09-03
The ReImageNet project has reannotated the entire ImageNet-1k validation set from scratch.
- All 50,000 images relabeled as multilabel, with 96,051 bounding boxes and new definitions for all 1,000 classes
- Every annotation is browsable online
- The accompanying paper, "Doomed to Re-Annotate, Forever: The ImageNet Story," warns anyone building a new classification/detection dataset: the team repeatedly thought they had it figured out and were always wrong
- Even benchmarks as established as ImageNet have far bigger annotation quality issues than assumed
Related event: ReImageNet: the full ImageNet-1k validation set relabeled(2 posts)→
More from Research
- SOCO benchmark debuts at ECCV 2026: 1M+ pairs probe how vision models grasp object structure — HirokatuKataoka · 2026-09-03
- Two-Author Model Tech Report Praised as Dense: Pretraining to Downstream — antoine_chaffin · 2026-09-03
- Recursive LM author clarifies: what experiments prove vs. what the work is about — CShorten30 · 2026-09-03
- Qwen-RobotManip: alignment before scale for robotic manipulation foundation models — rsasaki0109 · 2026-09-03
- Late-interaction doc embeddings shrink 255x to 6KB per page, keeping 95%+ accuracy — lateinteraction · 2026-09-03
- NeoMME-Retriever-260M tops sub-800M models on ViDoRe v3 with 0.523 nDCG@10 — lateinteraction · 2026-09-03