Research map / Label-efficient and robust learning / Self-supervised and contrastive
AutoNovel: Automatically Discovering and Learning Novel Visual Categories
Kai Han, Sylvestre-Alvise Rebuffi, Sébastien Ehrhardt, Andrea Vedaldi, Andrew Zisserman
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 BCE loss on pairwise pseudo labels (ablation) | AutoNovel · CIFAR-10 | ACC: 26.2 -> 90.4 | Improved |
| add MSE consistency constraint (ablation) | AutoNovel · CIFAR-10 | ACC: 82.6 -> 90.4 | Improved |
| add cross-entropy loss on labelled data (ablation) | AutoNovel · CIFAR-10 | ACC: 84.7 -> 90.4 | Improved |
| add self-supervised pretraining (ablation) | AutoNovel · SVHN | ACC: 72.9 -> 95.0 | Improved |
| add incremental learning scheme | AutoNovel · CIFAR-100 | ACC: 73.2 -> 75.2 | Improved |
| use ranking statistics instead of k-means (batch) for pairwise labels (ablation) | AutoNovel · CIFAR-10 | ACC: 42.9 -> 90.4 | Improved |
| use ranking statistics instead of nearest neighbor (ablation) | AutoNovel · SVHN | ACC: 78.2 -> 95.0 | Improved |
| use cosine similarity instead of ranking statistics (ablation) | AutoNovel · CIFAR-10 | ACC: 90.1 vs 90.4 | No clear effect |
| use soft ranking statistics (k=15) instead of hard (ablation) | AutoNovel · CIFAR-100 | ACC: 71.1 vs 73.2 | No clear effect |
| replace RotNet with MoCo v2 for self-supervised init (ablation) | AutoNovel · CIFAR-100 | ACC: 73.2 -> 62.5 | Hurt |
| use ImageNet pretrained model instead of self-supervision | AutoNovel · SVHN | ACC: 95.0 -> 40.2 | Hurt |
| use ImageNet pretrained model instead of self-supervision | AutoNovel · CIFAR-100 | ACC: 73.2 -> 87.1 | Improved |
Later papers that compared with this method
1 later accepted paper reports AutoNovel as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Automatically Discovering Novel Visual Categories With Adaptive Prototype Learning (TPAMI 2023 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Automatically Discovering and Learning New Visual Categories with Ranking Statistics (ICLR 2020 · similarity 0.94)
- Automatically Discovering Novel Visual Categories With Adaptive Prototype Learning (TPAMI 2023 · similarity 0.91)
- Learning to Discover Novel Visual Categories via Deep Transfer Clustering (ICCV 2019 · similarity 0.90)
- Semantic-Guided Novel Category Discovery (AAAI 2024 · similarity 0.88)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.86)
- Debiased Novel Category Discovering and Localization (AAAI 2024 · 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.
