Research map / Label-efficient and robust learning / Recognition and long-tail
The Met Dataset: Instance-level Recognition for Artworks
Nikolaos-Antonios Ypsilantis, Noa Garcia, Guangxing Han, Sarah Ibrahimi, Nanne Van Noord, Giorgos Tolias
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 |
|---|---|---|---|
| add PCA whitening and multi-scale (ablation) | R18IN kNN · Met dataset test | GAP: 3.7 -> 15.9 | Improved |
| use softmax confidence normalization (k=3 instead of k=1) (ablation) | R18IN kNN · Met dataset test | GAP: 2.9 -> 15.9 | Improved |
| include distractors in validation for tuning k and tau (ablation) | R18IN kNN · Met dataset test | GAP: 14.1 -> 15.9 | Improved |
| kNN on backbone instead of parametric classifier (AF loss) | R18IN DNet AF · Met dataset test | GAP: 16.9 -> 23.7 | Improved |
| ArcFace loss instead of cross-entropy | R18IN DNet · Met dataset test | GAP: 21.6 -> 23.7 | Improved |
| self-supervised SimSiam training on Met | R18IN kNN · Met dataset test | GAP: 15.9 -> 26.8 | Improved |
| add negatives to contrastive training (Con-Syn vs SimSiam) | R18IN · Met dataset test | GAP: 26.8 -> 30.4 | Improved |
| choose closest real positive instead of random real positive (ablation) | R18IN Con-Syn+Real · Met dataset test | GAP: 29.8 -> 32.5 | Improved |
| add random real positives to synthetic positives (ablation) | R18IN Con-Syn · Met dataset test | GAP: 30.4 -> 29.8 | Hurt |
| train on Met with contrastive loss starting from SWSL | R18SWSL · Met dataset test | GAP: 24.7 -> 36.1 | Improved |
| pretrain on SemArt art-attribute prediction instead of ImageNet | R50 kNN · Met dataset test | GAP: 22.2 -> 1.8 | Hurt |
| SWSL pretraining instead of ImageNet | R50 kNN · Met dataset test | GAP: 22.2 -> 30.4 | Improved |
Closest papers by meaning
- Museum Exhibit Identification Challenge for the Supervised Domain Adaptation and Beyond (ECCV 2018 · similarity 0.89)
- Discovering Visual Patterns in Art Collections With Spatially-Consistent Feature Learning (CVPR 2019 · similarity 0.86)
- Understanding Museum Exhibits using Vision-Language Reasoning (ICCV 2025 · similarity 0.85)
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (NeurIPS 2024 · similarity 0.84)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.84)
- APDDv2: Aesthetics of Paintings and Drawings Dataset with Artist Labeled Scores and Comments (NeurIPS 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.
