Research map / Label-efficient and robust learning / Recognition and long-tail
Delving into Inter-Image Invariance for Unsupervised Visual Representations
Jiahao Xie, Xiaohang Zhan, Ziwei Liu, Yew-Soon Ong, Chen Change Loy
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 inter-image invariance branch (InterCLR) | BYOL, 200 epochs · ImageNet linear | top-1 accuracy: 70.6 -> 73.5 | Improved |
| add inter-image invariance branch (InterCLR) | NPIDv2, 200 epochs · ImageNet linear | top-1 accuracy: 64.6 -> 65.7 | Improved |
| add inter-image invariance branch (InterCLR) | MoCov2, 200 epochs · ImageNet linear | top-1 accuracy: 67.5 -> 68.0 | Improved |
| add inter-image invariance branch (InterCLR) | BYOL, 1000 epochs · ImageNet linear | top-1 accuracy: 73.4 -> 74.5 | Improved |
| add inter-image invariance branch (InterCLR) | MoCov2, 200 epochs · VOC07 low-shot | mAP: 68.2 -> 70.7 | Improved |
| add inter-image invariance branch (InterCLR) | BYOL, 200 epochs · ImageNet 1% labels | top-5 accuracy: 76.8 -> 79.4 | Improved |
| add inter-image invariance branch (InterCLR) | BYOL, 200 epochs · VOC07+12 detection | AP50: 81.4 -> 82.7 | Improved |
| use online mini-batch k-means instead of offline global k-means for pseudo-labels (ablation) | NPIDv2-InterCLR · ImageNet / VOC07 (empirical study) | performance | Improved |
| use online labels instead of offline labels (training cost) (ablation) | Intra+Inter · 8 V100 GPUs, 200 epochs | training time: 6.4 days -> 5 days | Improved |
| use semi-hard negative sampling instead of hard negative sampling (ablation) | NPIDv2-InterCLR · VOC07 (empirical study) | mAP | Improved |
| use negative cosine margin m=-0.5 in inter-image branch instead of zero margin (ablation) | NPIDv2-InterCLR · ImageNet linear (empirical study) | top-1 accuracy | Improved |
| use positive cosine margin in intra-image branch (ablation) | NPIDv2 · ImageNet linear (empirical study) | top-1 accuracy | Improved |
Closest papers by meaning
- Unsupervised Representation Learning by Invariance Propagation (NeurIPS 2020 · similarity 0.90)
- Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning (CVPR 2022 · similarity 0.89)
- Contrastive Learning of Image Representations With Cross-Video Cycle-Consistency (ICCV 2021 · similarity 0.89)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.87)
- Linear Differential Vision Transformer: Learning Visual Contrasts via Pairwise Differentials (NeurIPS 2025 · similarity 0.86)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.86)
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.
