Research map / Label-efficient and robust learning / Semi-supervised learning
Invariant Information Clustering for Unsupervised Image Classification and Segmentation
Xu Ji, Joao F. Henriques, Andrea Vedaldi
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 auxiliary overclustering head (ablation) | IIC · STL10 (unsupervised clustering) | accuracy (%): 43.8 -> 59.6 | Improved |
| use 5 sub-heads instead of 1 (ablation) | IIC · STL10 (unsupervised clustering) | accuracy (%): 57.6 -> 59.6 | Improved |
| repeat each sample 5 times per batch instead of once (ablation) | IIC · STL10 (unsupervised clustering) | accuracy (%): 47.0 -> 59.6 | Improved |
| use the unlabelled data subset of STL10 (ablation) | IIC · STL10 (unsupervised clustering) | accuracy (%): 49.9 -> 59.6 | Improved |
| add entropy coefficient 1.5 to the loss | IIC · Potsdam-3 and COCO-Stuff-3 segmentation | pixel accuracy (%): 1.2%-3.2% | Improved |
| reduce labels by 90% (5000 to 500) in semi-supervised overclustering | IIC overclustering · STL10 (semi-supervised) | accuracy (%): ~10% drop | Hurt |
| use the auxiliary overclustering head with distractor-containing unlabelled data | IIC · STL10 (unsupervised clustering) | accuracy (%): 43.8 -> 59.6 | Improved |
Closest papers by meaning
- SCAN: Learning to Classify Images without Labels (ECCV 2020 · similarity 0.92)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.89)
- Unsupervised Few-Shot Image Classification by Learning Features into Clustering Space (ECCV 2022 · similarity 0.89)
- Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning (CVPR 2022 · similarity 0.89)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.87)
- Catalyst for Clustering-Based Unsupervised Object Re-identification: Feature Calibration (AAAI 2024 · similarity 0.87)
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.
