Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Image Clustering with Category-Style Representation
Junjie Zhao, Donghuan Lu, Kai Ma, Yu Zhang, Yefeng Zheng
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 mutual information loss to prior-only baseline (ablation) | DCCS · Fashion-MNIST | ACC: 0.618 -> 0.692 | Improved |
| add mutual information loss to prior-only baseline (ablation) | DCCS · CIFAR-10 | ACC: 0.213 -> 0.225 | Improved |
| add augmentation-invariant loss to prior-only baseline (ablation) | DCCS · Fashion-MNIST | ACC: 0.618 -> 0.725 | Improved |
| add augmentation-invariant loss to prior-only baseline (ablation) | DCCS · CIFAR-10 | ACC: 0.213 -> 0.645 | Improved |
| add mutual information loss to prior plus augmentation loss (ablation) | DCCS · Fashion-MNIST | ACC: 0.725 -> 0.756 | Improved |
| add mutual information loss to prior plus augmentation loss (ablation) | DCCS · CIFAR-10 | ACC: 0.645 -> 0.656 | Improved |
| use discriminator MI loss instead of decoder reconstruction loss (ablation) | DCCS · Fashion-MNIST | ACC: 0.732 -> 0.756 | Improved |
| use discriminator MI loss instead of decoder reconstruction loss (ablation) | DCCS · CIFAR-10 | ACC: 0.651 -> 0.656 | Improved |
| cluster with K-means on Z_c instead of argmax over Z_c (ablation) | DCCS · CIFAR-10 | ACC: 0.6556 -> 0.6513 | No clear effect |
| cluster with K-means on style part Z_s instead of category part (ablation) | DCCS · CIFAR-10 | ACC: 0.6556 -> 0.2940 | Hurt |
| convert CIFAR-10 images to grayscale (ablation) | DCCS · CIFAR-10 | ACC: 0.635 -> 0.656 | Improved |
| add Sobel filtering preprocessing (ablation) | DCCS · CIFAR-10 | ACC: 0.656 -> 0.652 | No clear effect |
Closest papers by meaning
- Deep Clustering via Joint Convolutional Autoencoder Embedding and Relative Entropy Minimization (ICCV 2017 · similarity 0.89)
- Clustering-friendly Representation Learning via Instance Discrimination and Feature Decorrelation (ICLR 2021 · similarity 0.88)
- Joint Unsupervised Learning of Deep Representations and Image Clusters (CVPR 2016 · similarity 0.88)
- Learning to Discover Novel Visual Categories via Deep Transfer Clustering (ICCV 2019 · similarity 0.88)
- Unsupervised Deep Embedding for Clustering Analysis (ICML 2016 · similarity 0.87)
- Deep Comprehensive Correlation Mining for Image Clustering (ICCV 2019 · 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.
